Identification of Sedimentary Microfacies with Well Logs: a New Approach of Indirect Classification

Multinomial logistic regression is a widely used classification approach in biology and sociology, while it can also have a high performance in sedimentary microfacies identification with wire-line logs. Taking K-successions... [ view full abstract ]

Multinomial logistic regression is a widely used classification approach in biology and sociology, while it can also have a high performance in sedimentary microfacies identification with wire-line logs. Taking K-successions of the H area of the Pearl River Mouth Basin as an example, we used two approaches to implement the microfacies identification. One is direct multinomial logistic regression, the other is an indirect approach which conducted the lithofacies classification with multinomial logistic regression first and then identified the microfacies based on previously estimated lithofacies. Both of the approaches were trained and examed by interpretations of an experienced geologist from real subsurface core data. The result show that the direct approach performs relatively poor with an total accuracy of 75% (some class even below 50%). While the indirect approach performs much better that with an overall accuracy near to 95% (none of the classes below 75%). This may because wire-line logs are more sensitive to lithologic features rather than sedimentary facies/microfacies. This indirect method is easy and reproducible and could be a robust way for analyzing sedimentary microfacies of horizontal wells, which have little core data or even are almost never cored, as long as core data is available for nearby vertical wells.