Hydrogeophysics Data for Machine Learning Hydrofacies Classification

Direct interpretation of complex and heterogeneous geological systems in terms of facies from individual tomograms remains a significant challenge because of noise in the measured data and non-uniqueness in geophysical inversion. We introduce a machine learning-based approach for hydrogeophysical image reconstruction using the Expectation-Maximization algorithm for joint classification of distinct hydrofacies from two or more independently inverted geophysical data sets.

To understand the impact of noise on facies discrimination, we design two synthetic models of hydrofacies with varying levels of complexity and heterogeneity. This synthetic study allows us to compare the classified image with the benchmark model for different noise scenarios. With field data in the Laramie range, WY, USA, we explore the effects of regularization on the hydrofacies classification.

The dataset consists of electrical resistivity and seismic velocities in both the field case and the synthetic case.