Automated feature classification and knowledge extraction from wireline geophysical observations: big data potential for offshore resources assessment

Scientific drilling of the volcanic ocean crust recovers cores and undertakes downholewireline logging. However, because core recovery rates are typically low (<30%), interpreting thewireline data is essential to gain a complete understanding of the stratigraphy. Ocean Drilling ProgramHole 1256D samples 1500 m of in situ upper oceanic crust and has both core-derived lithostratigraphyand electrofacies classification based on geological interpretations of continuous downhole FormationMicroScanner imagery. We propose an automatic quantitative identification of electrofacies usingDecision Trees. The cores and existing electrofacies classification provide training and verification ofthe automated classification. The identification of various classes is a challenging problem due tomissing data, vertical shifts, horizontal misalignments, and multiclass unbalanced problem with 2classes representing 50% of the data. Additionally, the structure of the same class changes with depthleading to large intra-class variations. Distinctive features for each class were identified by observationof images based on texture/shapes, and Decision Tree classifier was trained. Classification accuracyabove 90% was achieved for the 3-classes for electrofacies with high recovery rates. In case of 9-classes, accuracy above 60% was achieved for some classes, though some challenges are remaineddue to strongly overlapped classes. A detailed analysis of the big data used for training the classifierand its performance is described. Combined analysis of drill cores and wireline geophysical data fromscientific boreholes into volcanic rocks provides excellent training opportunities to develop automatedrock classification methods for complex geological terranes that are of increasing interest to thehydrocarbons industry.