Estimation of gas hydrate saturation in the Ulleung basin using seismic attributes and a neural network

Abstract

Among the unconventional natural resources, gas hydrates have recently received much attention as a promising potential energy source. To develop gas hydrates, their distribution and saturation should be estimated, preferentially at the initial stage of development. In most cases, the distribution of gas hydrates can be identified by using seismic indicators including a bottom simulating reflector (BSR) and chimney/column structures, which indirectly determine the presence of gas hydrate. However, these indicators can be used only when they appear on a seismic image. Because the saturation of gas hydrate is generally calculated by using well logs, the information is limited to the well location. To overcome these limitations, seismic impedance inversion and neural network methods can be used. Seismic inversion enables the identification of a gas hydrate reservoir even if seismic indicators do not exist, and a neural network makes it possible to predict the gas hydrate saturation in a region of interest away from the wells by combining well logging data and other attributes extracted from the seismic data. In this study, to estimate the distribution and saturation of gas hydrates that are broadly distributed in the Ulleung basin of the East Sea, seismic inversions such as acoustic impedance (AI), shear impedance (SI), and elastic impedance (El) were calculated, and then the seismic attributes (ratio of compressional wave velocity to shear wave velocity, Vp/Vs, and combinations of lame parameters, lambda rho and mu rho) that have unique features in hydrated sediments were extracted. Gas-hydrate-bearing sediments displayed high AI, high SI, high El (22.5 degrees), low Vp/ Vs ratio, high lambda rho, and high mu rho compared the surrounding sediments. The sediments containing free gas displayed low AI, low SI, low El (22.5 degrees), high Vp/Vs ratio, low lambda rho, and low mu rho due to the phase transition from gas hydrate to gas. By combining these findings, the distribution of gas hydrates was estimated even if seismic indicators were not present in the seismic profile. Using the extracted seismic attributes, as well as standard seismic attributes and three-phase Biot-type equation (TPBE)-derived saturation logs of gas hydrates at the wells which had a high correlation to the seismic attributes, the saturation of gas hydrates away from the wells could be estimated based on probabilistic neural network (PNN) predictions. To validate the predicted saturation, cross-validation of wells was undertaken. The average correlation coefficient between the predicted saturation and actual saturation logs at the UBGH-09 and UBGH2-10 wells was 82.6%. In addition, for the estimation of the saturation section of gas hydrate, a relatively high saturation region of gas hydrate corresponded well to the gas hydrate occurrence zone of each well.