-
Y =
0 + 1FD1 + 2FD2 + .... + 30FD30.
Leg 181 was the maiden voyage for the automated system now used to collect color reflectance data. At the first site of Leg 181, Site 1119, the Minolta spectrophotometer did not consistently make good contact with core material, which resulted in poor data quality data for Site 1119 and left the shipboard color reflectance data unusable. Adjustments were made in the settings of the spectrophotometer before data were collected at subsequent sites, and therefore data quality for the remaining cores is better (Carter, McCave, Richter, Carter, et al., 1999). Color reflectance data (laboratory data) were collected on cores from Hole 1119C at the ODP GCR with a Minolta CM-2002 spectrophotometer. This is the same model instrument used aboard ship. The software package used in the data collection process was provided by the manufacturer. A barium sulfate standard was used. Cores were covered with Glad plastic wrap before taking measurements to protect the optical components of the instrument from moisture and scratches and to avoid sample contamination. A glass slide was used to scrape mold growth from core surfaces. Measurements were taken on cores by careful visual inspection to ensure good contact with the core material and avoid confounders such as cracks, large shell fragments, pebbles, and pyrite oxidation halos.
Figure F1 is a comparative display of the ranges of color reflectance data (all wavelengths combined) for each hole. The maximum, 75th quartile, median, 25th quartile, and minimum values are displayed in a box and whisker format. There are obvious problems with the data. Cores from Site 1120 show out-of-range values for Holes 1120B and 1120C (reflectance >100%). A comparison with the measured carbonate summary statistics (Table T1) shows that these values are significantly higher at Site 1120 than at the other sites. These problematic data were not discarded because no other color reflectance data from this site are available. Because the measured carbonate values for Site 1120 are higher than those from other sites and because the range of color reflectance values is similar to data from other cores, we utilized these data to construct the model for Site 1120. Interquartile range values for color reflectance data from Site 1120 are similar to those of other cores recovered during this leg, which suggests the possibility that they preserve some of the variance in the data set, even though absolute values are not reliable.
Models to estimate carbonate concentrations were generated using the statistics package produced by the SAS Institute, Inc., and are regression models. Models for Sites 1124 and 1125, and Holes 1119C and 1123A used the following model form:
0 + 1FD1 + 2FD2 + .... + 30FD30.
The model for Site 1121 has the form
0 + 1R1 + 2FD4 + 3FD7 + 4FD21 + 5FD30.
The model for Site 1120 has the form
0 + 1R1 + 2FD2 + 3FD7 + 4FD21 + 5FD27.
Models for Sites 1122 and 1123 (all shipboard data) are of the form
0 + 1R1 + 2FD1 + 3FD2 + .... + 30FD30.
For all these models,
i = the coefficient (model parameter) estimated in the modeling process.
Development of the models proceeded empirically. The choice of first derivative transformation of the reflectance data was suggested by the work of Deaton and Balsam (1991) and the collective experience of the authors. The data were analyzed to minimize the covariance problem that is inherent in reflectance data. Variables were selected based on this study and various combinations of the least covariant factors were tried in order to choose empirically the most effective regression model. Adding the first untransformed reflectance variable (R1 = 400 nm) helped increase the R2 value for some of the models. An indication of how well each model describes the data can be ascertained by inspecting Table T2. Models that have low enough R2 values (<50%) to be of questionable value as prediction tools are the models for Sites 1122 and 1123. Several of the models account well for most of the variation in the data—namely, the models for Sites 1121 and 1125 and for Hole 1119C. Plots of measured carbonate vs. predicted carbonate fraction can be examined for every model in "Appendix C." Model parameter estimates for every model are also included in "Appendix D."
In an effort to get an indication of the relative data quality between shipboard data and laboratory data, some cores from Hole 1123A were also measured in the laboratory after the cruise. Figure F1 shows that the data range for laboratory data is not significantly different from that of the shipboard data. However, the R2 values for the models are significantly different. The model generated using the laboratory data is significantly better than that using the shipboard data. Both models use the same variables, the same model structure, and the same depth range over the cores. This suggests that there may be a problem with the shipboard data in this interval, making it more noisy, masking the true variance of the data, and hindering the ability of the model to reflect the true variability of the data.
