IDENTIFICATION OF LOG UNITS THROUGH VISUAL INTERPRETATION AND MULTIVARIATE STATISTICAL ANALYSIS

Log units obtained from LWD data at Leg 196 sites were defined through a combination of visual interpretation and multivariate statistical analysis. Log units were identified visually by examining the character of the gamma ray, density, neutron porosity, resistivity, and photoelectric effect curves. First-order log units are related primarily to lithology corresponding to changes in the log character. Second-order log units represent subtle changes in log data, which are possibly nonlithologic.

Multivariate statistical analysis provided objective confirmation of the log unit boundaries. The multivariate statistical analysis entailed the following steps:

1. Each logging curve was normalized by subtracting the mean and then dividing by the standard deviation. The resulting curves have a mean of zero and a standard deviation of 1.
2. Factors and factor loadings were calculated from the normal-sized curves using standard R-mode factor analysis procedures, with Kaiser Varimax factor rotation as described, for example, by Davis (1973). Each factor is simply a linear combination of the input variables weighted by the factor loadings; the factors can be visualized as a projection of n input variables (the normalized logging data) onto n linearly independent (uncorrelated) principal axes. Generally, for the Leg 196 LWD data sets, more than 75% of the variance observed in the input variables can be described by the first three factors.
3. The factors were then decimated to a 1-m depth interval using a finite-impulse-response, low-pass antialiasing filter to reduce the number of data points. This step was necessary because of computational limits of the software and hardware used for the next step.
4. Finally, a complete linkage hierarchical cluster analysis (using Euclidean norm; Davis, 1973) was performed on the three decimated factors that accounted for the greatest percentage of variance observed in the data. This allowed the identification of electrofacies, or log units, with distinct combinations of logging properties (e.g., Serra, 1986).

Factor analysis is a method of reducing the number of logs without losing important information. Cluster analysis of the three most important factors proved to be a useful and objective method of identifying significant first- and second-order log units.