3
cm) used for direct carbonate measurement and the location of the nearest
reflectance measurements.Calibration equations were developed for each site and for each lithostratigraphic unit (or subunits in the case of Site 1063) described on board the JR within each site. Carbonate distribution changes between stratigraphic units (Keigwin, Rio, Acton, et al., 1998): Unit I at Sites 1054 through 1062 on the Blake-Bahama Outer Rise (BBOR) (Fig. F1) consists of a rhythmic alternation between nannofossil-rich and clay-rich sediments and corresponds to the intensification of the 100-k.y. orbital forcing within the last 900 k.y.; Unit II and Unit III are dominated by terrigenous sediments. A similar lithostratigraphy is apparent between the Subunits IA, IB, and IC at the Bermuda Rise (BR) Site 1063 (Keigwin, Rio, Acton, et al., 1998) (Fig. F1). Additionally, opaline silica (10-30 wt%; Keigwin, Rio, Acton, et al., 1998) was detected in Subunit IA on the BR. However, the general range of variation for carbonate content is similar for all sites and units (i.e., 0-70 wt%; Fig. F2).
For Unit I (Subunit IA at Site 1063) we developed prediction equations for each site (Table T2; Fig. F2). However, because of the limited number of available carbonate measurements, general calibration equations for Unit II (including Subunit IB at Site 1063) and Unit III (and Subunit IC at Site 1063) included all available data from all sites (Table T2; Fig. F2). We also developed a regional general equation using all carbonate data (Table T2; Fig. F2).
The variance explained in
each equation is high, as indicated by the adjusted r2,
which ranges between 0.92 and 0.99. The root-mean-square error (RMSE) of
estimation is between 4% and 7%, which is within the range of previous estimates
using multiple linear regression on diffuse reflectance data (Mix et al., 1995;
Balsam and Deaton, 1996; Harris et al., 1997; Ortiz et al., 1999). The largest
errors occurred for equations developed using data from multiple sites and for
the regional equation, which also crosses lithologic boundaries (Table T2).
One source of variability is the presence of detrital carbonate in glacial
Atlantic sediment (e.g., Flood, 1978; Balsam and Williams, 1993). Analytical
techniques do not discriminate among various carbonates that show different
reflectance spectra, therefore increasing the uncertainty in the calibration.
Another source of error is the mismatch between the location of the samples
(i.e., 3
cm) used for direct carbonate measurement and the location of the nearest
reflectance measurements.
When compared with the 1-2 wt% errors that are typical for the direct carbonate measurement methods, the errors in our method are acceptable for the general extensive range of carbonate contents (i.e., 0-70 wt%) that characterize sedimentation at Leg 172 sites. Moreover, the relatively low RMSE for the general calibration equation indicates that regional sedimentation is remarkably uniform. However, the performance of each calibration equation with data sets other than the one for which it was developed shows that variation in mineralogical composition exists (Tables T3, T4). The deep sites on the BBOR (Sites 1058-1062) and Site 1063 on the BR are similar, with the calibration equation developed for each site performing well (i.e., r2 = >0.90 and RMSE = <7.5%) for the other sites within the group (Tables T3, T4). There is also a similarity between Sites 1054 and 1055 drilled on the Carolina Rise, suggesting a similar mineral matrix. The equations for the sites situated at intermediate depth on the BBOR (i.e., Sites 1056 and 1057) performed equally well for both adjacent shallower or deeper sites, indicating their transitional character, but the most robust equation is the regional one.
High reflectance over the entire visual spectral range is characteristic for calcium carbonate (e.g., Gaffey, 1986). This high reflectance can be more easily distinguished in the blue end of the spectrum (400-500 nm), where increased reflectance from iron oxyhydroxides is absent. The dominant terms within the calibration equations are generally first derivatives (or their squares) of the reflectance situated between 400 and 500 nm and/or the reflectance terms (or their squares) within the middle section of the spectrum (500-600 nm). A negative term in regressions, corresponding to a reflectance value between 400 and 420 nm, most probably represents a correction for the iron oxides' absorbance band at 400-450 nm (e.g., Morris et al., 1985).
