Geology differentiation of geophysical inversions using machine learning

Abstract

Summary Multiple geophysical methods are often employed to improve subsurface understanding, especially in areas with little a priori geological information. Therefore, quantitative methods for integrating multiple physical property models are fundamental to taking the interpretation further into geology di ff erentiation of distinct units. Hence, applications of machine learning are growing in geosciences due to its potential to integrate various sources of information. We evaluate the performance of density-, distribution-, centroid-, and correlation-based clustering methods in the identification of the three geologic units in density, susceptibility and conductivity models derived from a synthetic model, and show that correlation-based clustering gives the best results for geology di ff erentiation. We apply the method to physical property models recovered from field data over a copper deposit and the results show a good spatial correspondence with the known geology from drilling information, allowing the construction of a quasi-geology model.