Carbonate content was determined on selected samples from each site. At Sites 994, 995, and 997 carbonate content was determined for only the upper ~200 mbsf of each site because below this depth carbonate concentrations are very low and show very little fluctuation (Fig. 2; Table 1). Several generalizations can be extracted from the downhole patterns of changing carbonate content (Fig. 2). The highest carbonate values are present in upper Pleistocene sections. Not only are these late Pleistocene carbonate values high, but the frequency of carbonate variation also tends to be highest in the upper Pleistocene section. Early Pleistocene values are also high, but not as high as in the upper Pleistocene section, and they exhibit high frequency variations. These variations are well displayed in the lower Pleistocene sediments of Hole 994C and, if a more detailed stratigraphy can eventually be established, may prove to be cyclic (Fig. 2). The upper Pleistocene section in the three holes with long records (Holes 994C, 995A, and 997A) ranges from 35-40 m in length. The lower Pleistocene section, unlike the upper Pleistocene section, is much less consistent in length and is either eroded or compressed in Hole 997A.
Pliocene carbonate values in Holes 991A, 994C, 995A, and 997A differ from Pleistocene values because they are lower, generally around 10% or less, and show less high-frequency variation. In Holes 994C and 997A, late Pliocene carbonate values are somewhat higher and exhibit higher frequency variations than early Pliocene samples. In this respect, the late Pliocene carbonate values in these holes are similar to the Pleistocene carbonate values.
Red- to rose-colored, hematite-rich sediments on the continental margin off eastern North America have been described in several studies (Ericson et al., 1961; Heezen et al., 1966; Conolly et al., 1967; Schneider et al., 1967; Needham et al., 1969; Hollister and Heezen, 1972). The mapped distribution of these "brick-red and rose-colored lutites" was a major piece of evidence that documented the existence of contour currents, in particular the WBUC, which flows southward parallel to the contours of the continental rise of the western North Atlantic and has formed the huge sediment drift deposits of the Blake-Bahama Outer Ridge (Heezen et al., 1966). This reddish clay was derived from the erosion of Permian-Carboniferous rocks in the Canadian Maritime provinces and was transported to the marine environment by icebergs and proglacial streams during late glacial times. The presence of this unique, point-sourced sediment in high concentrations is easily determined visually by its distinctive reddish color. However, visual examination cannot identify low concentrations of hematite, which is sensitive to masking by other minerals. Barranco et al. (1989) employed a diffuse reflectance spectrophotometer to identify the presence of hematite in brick red lutites, especially at low concentrations. In particular, the height of the first-derivative peak at 565 nm appears to be especially sensitive to hematite in concentrations typical of these red sediments (Deaton and Balsam, 1991). First-derivative peak height for hematite is a function of both the hematite concentration and the mineral composition of the matrix containing the hematite. In a dark matrix, like that of the Leg 164 sediments, hematite can be identified at concentrations as low 0.03 wt%; whereas in a white matrix, hematite can be identified at concentrations as low as 0.01 wt%. Regardless of the matrix, diffuse reflectance spectrophotometry is an order of magnitude more sensitive to the presence of hematite than visual observation.
In most of the Leg 164 holes, the first-derivative peak height at 565 nm is very low except in the top 2 m, indicating that hematite is not a common sediment component except in the very uppermost Pleistocene and Holocene sections (Fig. 3, Fig. 4, Fig. 5). Holes 993A and 994C both show modest increases in hematite content in the top 0.2 m of the holes (Fig. 4). Although Holes 995A, 996E, and 997A all show slight increases at the very top of the hole, they also exhibit a substantial hematite peak between 1.2 and 1.9 m (Fig. 4). Shipboard spectral data gathered with the Minolta are more closely spaced, about every 0.2 m in Hole 994C and 0.1 m in Holes 995A and 997A (Fig. 5). Downhole variations in the first-derivative value at 565 nm exhibits patterns similar to those gathered with the Perkin-Elmer spectrophotometer where the samples are further apart. The main difference between the curves is the smaller variation present in the Minolta data as a result both of measuring wet samples and subsequent muting of the spectral signal and the presence of a 565-nm peak at ~5 m in Hole 994C that is not present in the Perkin-Elmer data. The peak at ~5 m probably is the result of the closer sampling interval for the Minolta data. It is important to note that even though this peak is further downhole than the 565-nm peaks in other holes, it probably is late Pleistocene or early Holocene in age, based on the thickness of the carbonate high near the top of the hole.
Both the Perkin-Elmer and Minolta results are consistent with the grain-size studies of Ledbetter and Balsam (1985), which showed that during the last 25,000 yr, the WBUC reached the shallowest depth between 14-5 ka, during the latest Pleistocene and early Holocene. In an analysis of cores from the Blake Outer Ridge, Haskell et al. (1991) also observed coarse silt mean grain sizes in the upper Pleistocene sediments, suggesting a shallower WBUC. The absence of hematite in the holes below about 2 m indicates that since the latest Pleistocene, the WBUC (or at least one arm of the WBUC) is the shallowest on the South Carolina margin and the crest of the Blake Outer Ridge as it has ever been.
Investigation of several important questions concerning the shipboard gathering of spectral data by Balsam et al. (1997) revealed that more thorough studies were required to answer these questions. Outstanding questions include concerns about the reliability of shipboard spectral data gathered from the wet, split core surfaces. This problem has been addressed in another paper (Balsam et al., 1998) and the results suggest that: (1) spectra derived from wet core samples are darker than dry samples and may contain less information about sample mineralogy and composition, (2) the degree of darkening depends on both water content and sediment grain size (darkening increases with increasing water content and decreasing grain size), (3) in sediments with a water content higher than about 5%, the reflectance decrease is greater at the red end of the spectrum than at the violet, thereby muting and distorting the spectral signal, (4) when sediment water content is less than 5%, spectral curves are somewhat darker than curves from dried sediment, but are of similar shape, and (5) more spectral information is contained in powdered samples that have been dried than in wet samples. These findings make it especially important to assess the reliability of shipboard spectral data because changing water content may affect not only reliability, but also the shape of spectral curves. Because changes in the shape of spectral curves can be related to changes in mineralogy, the presence of water has the potential to hamper mineralogic interpretation based on spectra. In the present study, we attempt to determine, what, if any, mineralogical information is lost during spectral analysis of wet cores, and if additional information is gained by analyzing dry, prepared samples.
Factor analysis was used to examine the structure of both the shipboard (wet) and shore-based (dry) spectral data; that is, to determine if the variability in the data is contained in the same covarying wavelengths. To answer these questions we prepared three data sets from measurements and samples down Holes 994C, 995A, and 997A because these holes recovered long continuous sections. First, we prepared a data set of all shipboard Minolta spectra from the three holes. This data set consisted of 6766 spectra after problem spectra had been deleted. We then prepared a second data set from core samples (which were ground, made into a slurry, and dried onto a glass microslide) and analyzed them with the Perkin-Elmer Lambda 6 spectrophotometer. This data set contains 1190 samples taken at ~1.5 m intervals (i.e., 1 sample per section) on cores from Holes 994C, 995A, and 997A. A third data set was constructed from a subset of the Minolta spectral measurements that correspond to the same intervals in the cores as the Perkin-Elmer samples in the second data set. (Note: we considered a shipboard measurement that was within 20 cm of a core sample to be at identical depths).
The Minolta data sets are limited to the visible region of the electromagnetic spectrum, 400-700 nm. The shore-based, Perkin-Elmer data set, on the other hand, contains spectral values from 250-850 nm. We expected the large Minolta data set to encompass the full range of sedimentologic variability in the three holes. The Perkin-Elmer data set, while more limited in sample spacing, has a larger spectral range and was expected to indicate the presence of a wider variety of minerals. Comparison of both these data sets to the reduced Minolta data set was intended to provide information on the effect of using fewer samples and a smaller wavelength range.
Untransformed percent reflectance curves are difficult to compare because they are smooth and change gradually. Typically, reflectance curves from marine sediments increase slowly from 400 to 700 nm and contain few peaks or valleys (e.g., Balsam and Deaton, 1991; Balsam and Wolhart, 1993; Balsam et al., 1997). One method frequently used to make spectra easier to analyze is to take the ratio of various color or wavelength bands (e.g., Mix et al., 1992, 1995). Ratios, however, fail to take into account that reflectance in a particular band can be influenced by any number of similar minerals. For example, both chlorite and glauconite would increase reflectance in the green wavelengths. Hence, ratios are not usually indicative of mineralogy. As noted by Barranco et al. (1989), Balsam and Deaton (1991), and Deaton and Balsam (1991), changes in the slope of the reflectance curve (that is, the first derivative of the percent reflectance curve) are more indicative of sediment mineralogy and composition. First-derivative values are high where the slope of the percent reflectance curve changes rapidly. We calculated first-derivative values at 10-nm sample intervals for both the shore-based and shipboard spectra. For the shipboard data, these calculations resulted in 30 values for each spectral analysis, each of which was then plotted at the midpoint of each 10-nm interval (i.e., the 400 to 410 value is plotted at 405, etc.) and are expressed as %/nm. For the shore-based data, 60 first-derivative values were calculated from 250-850 nm. Typically, first-derivative curves contain more peaks than reflectance curves (see Barranco et al., 1989; Balsam and Deaton, 1991, for examples).
One conclusion was immediately obvious from analysis of both the full and partial shipboard Minolta data sets. Our previous recommendation use only Glad Cling WrapTM brand polyethylene food wrap, as opposed to other plastic food wraps of various compositions (such as Saran WrapTM) to cover the wet cores during spectral measurements (Balsam et al., 1997), eliminates a substantial amount of noise and extends the useful wavelength range of the instrument. Balsam et al. (1997) noted that in cores covered with Saran WrapTM, some wavelengths were difficult to explain with a reasonable number of factors. Communality is a measure that describes how well a particular variable, in this case wavelength, is explained by the factors. Specifically, in the Leg 155 data set, the wavelengths from 405-435 nm and 695 nm had low communalities (<0.7) and could not be easily explained with factors that contained more than one percent of the variance. As Balsam et al. (1997) pointed out, the low communalities of these wavelengths indicated they were dominated by noise. For both the full and partial Leg 164 shipboard Minolta data sets the lowest prior communality (that is, communality before rotation) is 0.76. In the full Minolta data set, only three wavelengths had communalities less than 0.80, and, in the partial Minolta data set, only two wavelengths had communalities less than 0.78. Two of the wavelengths with low communalities were mainly at the red end of the spectrum, 685 nm and 695 nm, and in the full data set, 535 nm also had a low communality. Significantly, in the samples covered with Glad Cling WrapTM communalities at the violet end of the spectrum were all high. In contrast, violet wave-length communalities from the Leg 155 data covered with Saran WrapTM were low. The slightly lower communalities at the red end of the spectrum observed in the Leg 164 data set appear to be a function of the limits of resolution of the instrument, because we have observed similar communalities from data sets that were not covered with any type of plastic wrap. These data confirm our previous findings (Balsam et al., 1997) that the use of Glad Cling WrapTM reduces noises and improves the resolution of the Minolta spectra taken on wet cores. Thus, we continue to recommend that only Glad Cling WrapTM be used to cover wet cores during shipboard measurement.
Factor analysis of the full and reduced Minolta data sets using the principal factors method revealed that four factors could explain 97.8% and 97.6% percent of the variance, respectively. In both data sets this variance is explained in four factors; the factors are nearly identical in both data sets (Fig. 6, Fig. 7). These factors explain approximately 55%, 29%, 7%, and 6%, respectively. All factors greater than four explain less than one percent of the variance of the data set. The similarities of these factor analysis results is surprising, especially considering that the full Minolta shipboard data set contains about 5.5 times more samples than the Minolta data set reduced to match the Perkin-Elmer data set. One conclusion that can be drawn from these results is that broad sediment characterization can be accomplished with approximately one sample per section. Clearly, this strategy will miss events between the ~1.5-m sample, but should be sufficient to define long-term trends.
Factor interpretation is not straightforward. Balsam and Deaton (1991) presented a thorough discussion of factor interpretation as applied to spectra. Factors are interpreted by comparison to the data from which they were derived, in this case first-derivative values. For most factors, the factor pattern curves (Fig. 6, Fig. 7), which describe how important each wavelength is in each factor, should be similar to the first-derivative curve of a sediment component or mineral present in the sample. First-derivative curves for various minerals and sediment components have been published in Barranco et al. (1989), Balsam and Deaton (1991), Deaton and Balsam (1991), Balsam and Wolhart (1993), and Balsam et al. (1997). However, because factors are covarying combinations of wavelengths, they may include more than one mineral or sediment component, which makes them difficult to interpret. Two other methods are sometimes useful for interpreting factors. First, if the data set being analyzed is geographically distributed, the factors can be mapped and the maps compared to the distribution of a known sediment component. This is the method Balsam and Deaton (1991) used to identify the carbonate and organic factors, which are sediment components whose signal is muted (e.g., carbonate) or highly variable (e.g., organic matter). Second, factor scores, which describe how important each factor is in each sample, may be compared to known sediment components. Because marine sediments are complex mixtures of components and because spectra don't necessarily respond in proportion to a component's weight, this last method may produce misleading results. For example, 0.05 wt% hematite may have a major effect on spectra (Deaton and Balsam, 1991), but is beyond the analytical limits of other techniques commonly employed for mineral identification (e.g., XRD).
The four factors identified in the shipboard Minolta data from the Leg 164 holes (Fig. 6, Fig. 7) are interpreted as (1) carbonate, (2) either chlorite, glauconite, or a mixture of the two, (3) iron oxides, probably a mixture of hematite and goethite, and (4) organic matter. Factor 1 is interpreted as a carbonate factor because its shape is similar to that of the carbonate factor identified by Balsam and Deaton (1991). Based on comparison to first-derivative curves (Fig. 8), factor 2 is interpreted as chlorite, glauconite, or a mixture of chlorite and glauconite. Both chlorite and glauconite exhibit high first-derivative values between about 425 and 505 nm, the same range where factor 2 exhibits high values. In addition, like factor 2, both glauconite and ripidolite (Fig. 8) exhibit strong negative values from about 575 to 645 nm. The position of this valley extending into the red wavelengths suggests some clinochlore is present, but in the VIS wavelength range analyzed by the Minolta spectrophotometer, no clear determination between glauconite and the specific type of chlorite present is possible. We emphasize that all the samples shown on Figure 8 are natural samples and substantial compositional and spectral variability is likely.
Factor 3 is interpreted as a combination of iron oxides containing hematite, goethite, and possibly other oxides and oxyhydroxides. The first-derivative curve for hematite exhibits a peak at 575 nm, whereas goethite exhibits peaks at 535-545 nm and 435 nm (Fig. 9). When these two minerals occur mixed, as is typical in nature, their spectral curves are additive and have the net effects of shifting the hematite peak toward shorter wavelengths, broadening the area under the peak and producing a shoulder at 525-535 nm.
Characterizing organic material spectrally is difficult because the composition and color of organic material is highly variable. Balsam and Deaton (1991) and Balsam and Wolhart (1993) previously identified first-derivative values that increase toward the red end of the spectrum as typical organic material. Our factor 4, however, exhibits high first-derivative values at the violet end and low values at the red end of the spectrum. The precise pattern produced by organic matter is a function of the type of organic material and its concentration. Previous models of sediment organic material included a slight amount of lignite, which causes absorption through most of the VIS region of the spectrum (Balsam and Wolhart, 1993). In these Leg 164 samples, there appears to be very little refractory organic material (lignite, for example), and the factor pattern curve (Fig. 6, Fig. 7) is similar to the first-derivative curve produced by combining organic matter (1 wt% cow manure) with calcite (Fig. 10). It is also important to note that low-weight percent organic matter, because of its light weight, may represent a substantially higher volume percentage.
To further assess the quality of the shipboard Minolta data, we took the shore-based Perkin-Elmer data and factor analyzed it from 400-700 nm, the same wavelength range as the Minolta data. The Perkin-Elmer data differs from the Minolta in that (1) the Perkin-Elmer spectrophotometer has a better signal-to-noise ratio and records data at a 1-nm interval, as opposed to the 10-nm interval of the Minolta, (2) the samples were not covered during analysis, and (3) the samples were ground, made into a slurry, and allowed to dry on a glass microslide. These differences suggest the Perkin-Elmer should produce higher quality data. This assumption was verified by the high communalities, all >0.9, of the first-derivative wavelengths, which suggest that all wavelengths were well explained by the factor analysis. Unlike the shipboard Minolta data, which contains four factors, the Perkin-Elmer data contains five factors explaining more than one percent (Fig. 11). These five factors explain ~61%, 25%, 10%, 2%, and 1% respectively. Although the number of factors and their relatively importance differs, the four Minolta factors are easily recognizable in the Perkin-Elmer data set (Fig. 11). Factor 1, carbonate, is similar in both the Minolta and Perkin-Elmer data sets. Factor 2 in the Minolta data set is similar, but different in some important aspects, to factor 2 in the Perkin-Elmer data set and is discussed below. Factor 3 in the Minolta data set, iron oxides, is similar to factor 4 in the Perkin-Elmer data set, and the Minolta factor 4, organic matter, is similar to factor 3 in the Perkin-Elmer data set. The Perkin-Elmer data set clarifies the interpretation of the iron-oxide factor which, based on the 565-nm peak, is clearly dominated by hematite. Deaton and Balsam (1991) indicate that as hematite content decreases, the position of the main hematite peak shifts toward shorter wavelengths, which suggests a hematite content of <1% in these samples. The Perkin-Elmer data also clarify the interpretation of the chlorite-glauconite factor. In the 400-700-nm Perkin-Elmer data, the peak in factor 2 shifts to shorter wavelengths, which suggests the presence of clinochlore. In addition, the fifth factor appears to be either glauconite or ripidolite, which, because of their similar spectral patterns in the VIS (Fig. 8), cannot be distinguished. In conclusion, although the Perkin-Elmer is clearly better at discriminating between different minerals, the shipboard data taken with the Minolta do a remarkably good job distinguishing different sediment components.
To determine if additional compositional information could be obtained by analyzing a broader spectral range, we factor analyzed the shore-based Perkin-Elmer data from 250-850 nm (Fig. 12). In factor analyzing this large data set, 60 first-derivative variables by 1190 samples, we deleted first-derivative wavelengths with communalities less than 0.75 and stopped factor extraction when the variance explained was less than 1%. In addition, if a factor produced high-frequency variation (that is, noise) it was not included in the analysis. Using these criteria, our factor analysis included wavelengths from 265-725 nm and extracted six factors explaining 97% of the cumulative variance. Compared to the wavelength range available, the reduced spectral range used in the factor analysis suggests that the portion of the spectrum from 730-850 nm is not particularly useful for determining the composition of these marine sediments. Balsam et al. (1997) came to a similar conclusion and suggested that spectral instruments designed to scan marine core surfaces should be limited to wavelengths shorter than about 730 nm. The results of the present study support this conclusion.
The six factors (Fig. 12) extracted from analyzing the Perkin-Elmer data from 265-725 nm are interpreted as iron oxides (factor 1), illite (factor 2), chlorite and glauconite (factor 3), organic material (factor 4), montmorillonite(?) (factor 5), and carbonate (factor 6). Of these factors, the illite and montmorillonite factors require further discussion. Factor 2 exhibits generally high values from 300-400 nm; a similar pattern is exhibited by the first-derivative curve for illite (Fig. 13). We are aware of no other mineral that consistently exhibits high values throughout this range. It is important to note that the significant peaks for interpreting illite's spectral signal are present in the NUV and hence not available to instruments like the Minolta that analyze only the VIS. Factor 5 appears to be montmorillonite, specifically Fe-rich montmorillonite or nontronite (Fig. 14). Factor 5 (Fig. 12) and the first-derivative curve for nontronite (Fig. 14) both exhibit 685- and 535-nm peaks. A nontronite factor would not be expected to exhibit the 475-nm peak found in the nontronite curve, because that peak overlaps with the chlorite-glauconite factor. However, the peak at 395-405 nm is probably also indicative of nontronite. It is important to note that the addition of iron substantially changes the spectral pattern of montmorillonite (Fig. 14) and thus demonstrates how difficult interpretation natural materials can be. These data suggest that the extended wavelength range provides additional information.