Revealing Multiple Geological Scenarios Through Unsupervised Clustering of Posterior Realizations from Reflection Seismic Inversion

Abstract

In this study, we analyze 26000 posterior realizations obtained through Monte Carlo sampling from the posterior distribution of a reflection seismic inverse problem and show that the posterior realizations cluster around multimodal peaks. This problem is based on a seismic trace recorded in the southern part of Jutland, Denmark. Prior information is based on observations of lithology sequences of the geology in the area, and the multimodal modes in the posterior realizations will hence represent different geological scenarios. In order to uncover the multimodal nature of the posterior distribution, grouping of posterior realizations is done using an unsupervised clustering technique, namely the K-means clustering algorithms. In order to quantify the choice on the number of clusters in the realizations, the gap-statistic-method is used. The clustering method is applied on both categorical model parameters representing lithological units, and on the continuous parameters representing the acoustic impedance. These techniques allow quantifying the probability of the different possible geological scenarios that are consistent with the seismic and geological observations. Results demonstrate that the cluster characteristics are significantly dependent on the types of parameters considered. If the goal of the inversion is to identify different geological scenarios using a parameterization based on lithological units is more informative than a parameterization based on acoustic impedance.