Probabilistic Estimation of Seismically Thin-Layer Thicknesses with Application to Evaporite Formations.

The identification of potassium (K) and magnesium (Mg) salts prior to the well drilling is a key factor to avoid washouts, closing pipes, fluid loss damage, and borehole collapse. The Bayesian classification combines the outcomes from statistical rock physics and seismic inversion, providing the spatial occurrence of the most-probable salt types. It serves as a facies identifier of Mg–K-rich salts (bittern salts) before drilling. Nevertheless, the most-probable classification is limited to the seismic resolution which may underestimate seismically thin-layer thicknesses. Along with the most-probable facies, the Bayesian classification renders the facies probability volume. We demonstrate that the facies probability and facies-specific total thickness highly correlate to each other even under the threshold of seismic resolution. Thus, we employ the bittern-salts probability volume to predict thin-bed bittern-salts thickness in undrilled locations. To capture the variability of the seismic estimation, we resort to Monte Carlo-assisted simulations of wells that emulate the layering patterns of a site-specific deposition environment. These simulations are crucial to assist the estimation of the joint probability density function between the facies volume and the total thickness. Therefore, given the facies probability, the joint probability density function enables us to derive the conditional expectation and percentiles of thin-bed thicknesses. Furthermore, this paper proposes a method to quantify the negative influence of seismic noise in the estimation of thin-bed thicknesses. The blind well confirms the consistency of this technique to unfold the uncertainty in the seismic predictability of thin layers. We argue that this procedure is extendable to other facies. [ABSTRACT FROM AUTHOR]