In high risk, high cost environments, refining advanced technologies for accurate and reliable reservoir predictions are paramount. Advanced seismic technologies for development and production decisions in offshore areas have increased dramatically in recent years with the use of AVO attributes and seismic inversions. By utilizing the available well control throughout a reservoir, AVO attributes can be calibrated using artificial neural networks to generate higher fidelity AVO attributes that assist in refining the reservoir description.
Amplitude variation with offset (AVO) techniques are used by exploration, development and production teams to assist hydrocarbon identification in clastic depositional settings. While exploration groups tend to use AVO attributes for detection and risk quantification, exploitation and production groups use AVO attributes for reservoir characterization and even fluid-front monitoring. Accurate geoscience and engineering reservoir characterization (parameterization) improve prediction of hydrocarbon reserves and reservoir production. It is therefore essential to understand which seismic attributes will best contribute to the characterization of the reservoir. This paper focuses on the accuracy of AVO attributes commonly used in reservoir characterization. In particular, the effectiveness of interceptgradient, ???, and elastic impedance AVO attributes, and their ability to accurately predict reservoir extent are presented. Derivative AVO attribute volumes that are calibrated to well data are also compared.
This paper examines methodology differences between the various AVO attributes and, more importantly, compares the final reservoir description predicted by these attributes. Please note that looking for the best AVO attribute for reservoir characterization does not mean that this attribute will be the sole seismological contribution to the reservoir parameterization for reservoir simulations. Rather it is a method to determine the best AVO input (if any) to accompany other geophysical and geologic inputs to the modeling.
A data model with variable reservoir thicknesses was constructed using well log data from the middle Miocene section on the northern continental shelf of the Gulf of Mexico. A thin blocky sand encased by shales was selected as the reservoir. This sand, when gas-saturated, can be categorized as a class 2 AVO reservoir or, in other words, a gas reservoir with a higher velocity and lower density than surrounding shales. The model was constructed by seeding the reservoir with individual full-elastic wave equation synthetic seismograms with a selected cell thickness and fluid type. The resulting 35 in-line by 50 cross-line model contains two sands that vary in thickness from zero to 11 m at a depth of 2700 m (2.2 s). The model is essentially devoid of structure. A zero-phase 4/8-24\48 Hz wavelet was used, and data computed to a maximum offset of 4900 m. Velocity information for angle estimation and AVO gradient calculation came from a smoothed version of the initial sonic log; a robust fit was used to eliminate outlying points in the gradient measurements. (Since full-elastic wave equation synthetics were generated, "noise" from multiples and converted waves is present which can result in outliers in gradient calculations.)