Applying spectral decomposition to seismic facies clustering with unsupervised machine learning

Abstract

Seismic facies analysis, essential for subsurface geological exploration, has traditionally challenged the ability to capture subtle variations in complex stratigraphic environments. This study uses spectral decomposition and unsupervised machine learning, specifically the Kohonen Self-Organizing Map, to improve the identification of detailed seismic facies. Spectral decomposition enables frequency-based seismic data analysis, capturing intricate geological features often missed by traditional methods. The Continuous Wavelet Transform was applied to decompose seismic signals, and the resulting frequency components were clustered using a Self-Organizing Map to classify seismic facies. This paper validated this approach using seismic data from the South Caspian Basin. The results successfully identified channel systems and facies boundaries, enhancing their delineation and enabling a more accurate interpretation of channel systems and their internal variability. This automated methodology offers valuable insights for reservoir characterization and hydrocarbon exploration, potentially reducing exploration risks and enhancing resource estimation

Keywords

• seismic facies
• spectral decomposition
• self-organizing map
• continuous wavelet transform
• unsupervised learning
• machine learning