Investigating the predictive performance of Gaussian process regression in marine controlled-source electromagnetic modelling

• Main Authors: Mohd Aris, Muhammad Naeim, Nagaratnam, Shalini, Mohd Noh, Khairul Arifin, Daud, Hanita, Maamor, Nahamizun
• Format: Conference Proceeding
• Language: English
• Subjects: Covariance; Error analysis; Gaussian process; Geophysical methods; Marine environment; Mean square errors; Numerical methods; Performance prediction; Predictive control; Regression models; Root-mean-square errors
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0227908

This work presents a study on investigating the predictive performance of Gaussian process (GP) regression in modelling marine controlled-source electromagnetic (CSEM) data. Marine CSEM application is a surveying geophysical technique used to detect high-resistive body in marine environment. This application employs complicated numerical methods to model the electromagnetic (EM) profiles. Here, GP is utilized as the surrogate model to the complex marine CSEM model by treating the deterministic function as a process of random realization. The predictive performance of GP modelling relies on the covariance function used in the regression. In this work, four different types of covariance function, such as Matern 3/2, Matern 5/2, rational quadratic, and squared exponential were investigated in modelling two different sets of EM data with different transmission frequency, depth and resistivity of hydrocarbon. GP models with these covariance functions were developed at testing input specifications by utilizing data obtained from the marine CSEM computer experiment. Mean square error (MSE) and root mean square error (RMSE) between each GP model with different covariance function in every set and the corresponding true EM values were then computed and compared. The magnitude versus offset plot shows that GP regression is able to fit and compute the EM responses very well. Based on the error measurements, GP regression with squared exponential is the most suitable model in predicting and modelling the EM responses in marine CSEM application.