5.1. Can We Use Gamma Ray Data as a Proxy for Chemical Composition Downhole?
 The average K2O contents per core, calculated with the core-derived proportions are more scattered than those calculated using the log-derived proportions (Figure 9a). The latter plot closely on a one to one line, indicating a close agreement between the gamma ray data and K2O contents calculated from the samples using the new proportions, adjusted for missing intervals (Figure 9a). The average concentrations per core, calculated using the MST-NGR data are also more scattered than those calculated using log-derived proportions. These concentrations plot close to the contents calculated using the core-derived proportions, although usually closer to the 1:1 line (Figure 9a). This result can be expected, as the MST-NGR data, measured on the recovered cores, also do not take into account the missing intervals. It demonstrates, however, the reliability of MST-NGR data and their close agreement with the chemical compositions of the samples. It also shows an averaging effect related to this type of data and comparable to the one observed for logging data. The volume of core analysed with the MST-NGR tool will be larger than the discrete sample measured by chemistry and therefore variations in the chemical compositions reported by the MST-NGR data will be less extreme.
Figure 9. Plot of the average (a) K2O (wt.%) content per core derived from the geochemical analyses versus the K2O (wt.%) content per core derived from the gamma ray logging tool measurements and (b) U (ppm) content per core derived from the geochemical analyses versus the U (ppm) content per core derived from the gamma ray logging tool measurements. Each data point is representative of a core. The blue squares were calculated using the rock types proportions derived from the recovered cores and the pink circles were calculated using the lithology proportions derived from the core-log integration study. The yellow triangles reported on Figure 9a were calculated using the core-derived proportions and the MST-NGR data collected during Leg 185.
Download figure to PowerPoint
 Most of the K2O contents calculated with the core-derived proportions are lower than the content reported by the wire line logging tool (Figure 9a). This is related to two main effects: (a) the logging tool overestimates the K2O contents or (b) using the core-derived proportions results in a bias against K2O-rich rock types. The second explanation is likely as potassium will be more abundant in the most altered part of the basement section as well as in breccia and interpillow material [Alt et al., 1998] and these brittle rock types are preferentially lost during drilling process [Brewer et al., 1998, 1999]. It follows that average K2O content of the cores will be underestimated if the core-derived proportions of the different rock types are used.
 It should be also noted, that some cores show average K2O contents derived from the chemical analyses that are higher than the K2O content reported by the logging tool (Figure 9a). This can be observed on both sets of calculations, using either core-derived or log-derived proportions and cannot be related to a bias in the lithological proportions. It therefore must reflect a difference in the absolute K2O content measured by the gamma ray tool and the chemical analyses. In this case, the question is whether the gamma ray logging tool underestimate the potassium content in the section or if the K2O contents measured on discrete samples and used for a specific lithology are an overestimation.
 Some samples, mainly breccia, display higher K2O contents than measured by the logging tool, although the gamma ray curve does show peaks for these intervals (Figure 10, samples 185-801C-15R1-19-23 and 16R2-27-31, see also Figure 6). These samples comprise a matrix composed of highly altered basalt, carbonates, celadonite and clay minerals, preferentially enriched in K2O, and a small proportion of basalt clasts. In contrast, the third breccia sample (Figure 9, 185-801C-16R5-62-65) is mainly composed of basaltic clasts (90%) with only a small amount of matrix and its K2O content is in close agreement with the tool measurement. This difference can either be related to the logging tool itself, which measures the gamma ray counts over a larger volume than the analysed sample or to the samples themselves. Some breccia and pillow basalts samples have extreme compositions and are probably not highly representative of the overall composition of these rock types in the basement section. Other samples, in which basalt and matrix proportions are less extreme, are probably more representative of the overall composition (Figure 10). This can explain most of the differences observed between the chemical analyses and the gamma ray data for the pillow basalts, sediments and breccias (Figure 6). Despite uncertainties related to the radius of investigation of the tool and its averaging effect, this indicates that the gamma ray tool is well calibrated in basement sections and logging data can be used to accurately estimate the overall K2O content in a borehole.
Figure 10. Plot of K2O contents of discrete samples and gamma ray data versus depth for the interval from 600 to 640 mbsf.
Download figure to PowerPoint
 The same features, as for K2O, can be observed on the U calculations (Figure 9b). The average compositions per core calculated using core-derived proportions are slightly more scattered than those calculated using log-derived proportions. The latter also plot close to the one to one line (Figure 9b) and also demonstrate the reliability of the method. The concentrations calculated using the core-derived proportions are also slightly lower than the others and indicate the same bias toward U-rich lithologies, as observed on the K2O calculations. The bias toward high concentrations observed for K2O is not observed for U (Figure 9b, see also Figure 7). A ready explanation to this is found in the small contrasts in U concentrations between the different lithologies. The range of U concentrations in all lithologies are comparable and no extreme compositions are found in contrast to the K2O data. The sample data are therefore probably more representative of the different lithologies. We can conclude that the gamma ray tool is also fairly well calibrated and that U data can be used as a further constraint on the chemical composition of a basement section.
 Despite relatively large errors on the calculations (up to 20%), mainly dependent on our ability to interpret the logging data and to recognise the missing lithologies, our results show that this method can be used with confidence to estimate the K2O and U contents of a basement section. Furthermore, we demonstrate that the logging data are well calibrated and these results can be directly used to estimate the composition of the altered crust in term of K and U.
 This method also highlights the difficulty of choosing representative samples to use in a “composite” sample method. A major approach of ODP Leg 185 is to use such samples, which are a physical mixture of different rock types [Staudigel et al., 1996; Plank et al., 2000] to estimate the overall composition of the basement recycled in the Marianna subduction zone. The principal difficulty when preparing these composites will be first to choose the appropriate samples and then to choose in which proportions to mix them. If the samples are not representative of each rock type, but have extreme compositions, then neither of the core-derived or log-derived proportions will be representative. This is particularly critical for high K lithologies like breccia samples. Using the core-derived proportions may result in an underestimation of the concentrations but the use of log-derived proportions may also entrain an overestimation of the concentrations in the basement. However, the core-log-geochemical integration method is only applicable, so far, to two chemical elements, K and U in contrast to the composite sample approach which allow to analyse all the chemical elements as well as isotope compositions. A combination of these two methods is required to fully estimate the composition of the oceanic crust. This can be undertaken if samples are chosen to be representative of an average rock type by checking their composition against logging data and if enough samples of each rock type are analysed such that the range in composition is accurately represented. In this case, the log-derived proportions should be used to prepare the composite.