Estimating Marine CSEM Responses Using Gaussian Process Regression Based on Synthetic Models

Seabed logging (SBL) is an application of controlled-source electromagnetic (CSEM) waves to discover marine hydrocarbon-filled reservoirs based on the resistivity contrast of subsurface underneath the seabed. Current practice for processing marine CSEM responses utilizes meshes-based algorithms. The ad hoc algorithms require high computational time to solve the integrals and linear equations. Therefore, this synthetic-based study proposes Gaussian process regression (GPR) to estimate marine CSEM responses at various resistivities of target layer. Synthetic multifrequency SBL responses with target depth of 500 m from the seabed are modelled by finite element (FE) method using computer simulation technology (CST) software. As the prior information to the GPR, the target layer is parameterized with resistivity of 30–510 Ωm with an increment of 60 Ωm. By using MATLAB software, a two-dimensional (2D) GPR model is developed to estimate the marine CSEM responses at unobserved resistivities. For the validation, the mean absolute deviation (MAD), mean squared error (MSE) and root mean squared error (RMSE) between the 2D GP model and the CST outputs (i.e., true values) at the unobserved resistivities are calculated. The computational time for evaluating the marine CSEM using GPR and FE are computed and compared. The resulting error measurements and the computational time revealed that GPR can estimate the marine CSEM responses efficiently and at par to the current methods.