Modelling 1-D synthetic seabed logging data for thin hydrocarbon detection: An application of Gaussian process

• Main Authors: Aris, M.N.M., Daud, H., Noh, K.A.M., Dass, S.C., Mukhtar, S.M.
• Format: Conference or Workshop Item
• Published: American Institute of Physics Inc. 2020
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85094605486&doi=10.1063%2f5.0018105&partnerID=40&md5=f3238d7559f547ac03f16df4feebe930; http://eprints.utp.edu.my/29872/
• Institution: Universiti Teknologi Petronas

Seabed Logging (SBL) is a technique that employs high-powered electric dipole source to emit electromagnetic (EM) signal to detect hydrocarbon (HC) reservoirs beneath the seabed. This application is based on electrical resistivity contrasts between target reservoirs and its surrounding. SBL analysis can become a challenging task when the target reservoirs are thin, and the contrasts of resistivity are not very significant. As HC reservoirs are getting thinner, target responses and reference (non-HC) responses are difficult to be distinguished. Addressing this problem, we propose a simple statistical method, Gaussian Process (GP), to model one-dimensional (1-D) SBL data with uncertainties quantification. In this paper, Computer Simulation Technology (CST) software was used to replicate SBL models with five different thicknesses of HC. Some characteristics of the SBL models such as seawater depth, reservoir thickness and reservoir depth were imitated as the case study of Troll West Oil Province, North Sea. We developed 1-D forward GP model for all the SBL responses. Both modelled responses, target and reference, were compared and mean percentage differences between the responses were then calculated. For every comparison, confidence bars for each modelled response were observed to confirm the existence of thin HC. For model validation, root mean square errors (RMSEs) between modelled and generated (CST software) data were calculated. The confidence intervals revealed that the target and reference responses are distinguishable for all HC thicknesses, and the calculated RMSEs showed that GP is reliable to be applied in SBL data to provide uncertainties quantification. © 2020 Author(s).