2019 AAPG Annual Convention and Exhibition:

Datapages, Inc.Print this page

Machine Learning Multi-Attribute Analysis for Gas Hydrate Identification


Gas hydrates that exist in the subsurface are often difficult to image with reflection seismic data if the seismic data lacks a strong bottom simulating reflector (BSR). In these cases, the imaging and detection of the gas hydrate stability zone (GHSZ) becomes particularly difficult, as hydrate detection relies heavily on the BSR, gas chimneys, or pockmarks on the seafloor. To address and understand these imaging complications, an unsupervised machine learning multi-attribute analysis is performed on 2D seismic data in the Pegasus Basin in New Zealand where the BSR is not continuously or clearly imaged. The analysis uses principal component analysis (PCA) methods, applied to a selected set of seismic attributes to identify meaningful combinations of attributes that provide insight into the seismic data. The PCA results in key attribute combinations that are difficult to interpret due to their multi-dimensional nature. To aid in this visualization, self-organizing maps (SOMs) are employed. Rock physics analysis has demonstrated that the inclusion of methane gas hydrates in the pore space results in a slightly increasing amplitude at the base of the gas hydrate zone, regardless of the fluid (brine or gas) in the pore space below the hydrates. This increasing amplitude is quite weak in strength, particularly in shale rich lithologies containing brine. In these scenarios, a BSR is not typically observed in the seismic data, even though gas hydrates do exist in the subsurface. To enhance the detection of the presence of gas hydrates, the multi-attribute analysis is performed with a series of seismic attributes that are capable of detecting the minute changes in the seismic waveform due to the presence of gas hydrates. The successful attributes are those that are sensitive to attenuation, frequency, and small amplitude anomalies. In this case, instantaneous attributes that detect changes in the frequency and phase tend to cluster together in the PCA to reveal the interface at the base of the GHSZ. Individually, some of these attributes have minimal success in identifying the seismically invisible hydrate. However, employing a multi-attribute analysis provides clearer insight into the identification of hydrate zones.