Lithological control on gas hydrate saturation as revealed by signal classification of NMR logging data

Abstract

AbstractIn this paper, nuclear magnetic resonance (NMR) downhole logging data are analyzed with a new strategy to study gas hydrate‐bearing sediments in the Mackenzie Delta (NW Canada). In NMR logging, transverse relaxation time (T2) distribution curves are usually used to determine single‐valued parameters such as apparent total porosity or hydrocarbon saturation. Our approach analyzes the entire T2 distribution curves as quasi‐continuous signals to characterize the rock formation. We apply self‐organizing maps, a neural network clustering technique, to subdivide the data set of NMR curves into classes with a similar and distinctive signal shape. The method includes (1) preparation of data vectors, (2) unsupervised learning, (3) cluster definition, and (4) classification and depth mapping of all NMR signals. Each signal class thus represents a specific pore size distribution which can be interpreted in terms of distinct lithologies and reservoir types. A key step in the interpretation strategy is to reconcile the NMR classes with other log data not considered in the clustering analysis, such as gamma ray, hydrate saturation, and other logs. Our results defined six main lithologies within the target zone. Gas hydrate layers were recognized by their low signal amplitudes for all relaxation times. Most importantly, two subtypes of hydrate‐bearing shaly sands were identified. They show distinct NMR signals and differ in hydrate saturation and gamma ray values. An inverse linear relationship between hydrate saturation and clay content was concluded. Finally, we infer that the gas hydrate is not grain coating, but rather, pore filling with matrix support is the preferred growth habit model for the studied formation.