Machine learning-based seismic characterization of deepwater turbidites in the Dangerous Grounds area, Northwest Sabah, offshore Malaysia

Abstract

Seismic interpretation is a critical aspect of hydrocarbon exploration, where geoscientists often struggle to accurately recognize patterns and anomalies in large datasets. Machine learning techniques offer a promising solution by allowing for the quick and accurate analysis of multiple and large-size seismic volumes. This study leverages seismic facies analysis, seismic attribute analysis, and supervised machine learning to identify and characterize turbidite deposits in the Dangerous Grounds region, an underexplored area recently revealed by high-resolution broadband seismic data. Through seismic stratigraphy, two distinct phases of turbidite deposition were identified: a lower unit showing higher amplitude and signs of faulting effect, and an upper, present-day unit characterized by lower amplitude and continuous reflectors. The attribute expression of these turbidites shows strong amplitude response, high relative acoustic impedance, and high gray-level co-occurrence matrix entropy emphasizing their distinctiveness from surrounding facies, with variations in reflector continuity and spectral decomposition providing further insight into their depositional processes and sediment characteristics. By applying nine machine learning classifiers with twenty seismic attributes as input, this study achieved over 99% accuracy in distinguishing turbidite facies from background, with the neural network, random forest, K-nearest neighbors, decision tree, and support vector machine exhibiting optimal performance. The study contributes significantly to the regional understanding of turbidite deposits through detailed machine learning-aided seismic characterization. It underscores the value of integrating domain knowledge with machine learning techniques in enhancing subsurface interpretations, offering a comprehensive methodology for seismic facies analysis in similarly complex and underexplored regions.