ABSTRACT

The recently acquired Cocuite 3-D survey from the Veracruz Basin, Mexico, provides a unique opportunity to study fluid- and rock-property relationships in gas-bearing deepwater sandstones. The gas-producing intervals consist of sinuous sandy channels and sandy turbidite lobes. These and other reservoir geometries were successfully imaged using conventional interval and windowed attribute-extraction techniques. Although numerous direct hydrocarbon indicators (DHI's) were delineated and mapped from several stratigraphic units, future success is dependent on identifying anomalies that represent high gas saturation.

Inversion and neural-network training models were developed to improve the tie between amplitude and gas occurrence. Shallow buried sands are associated with lower well log impedance, and gas occurrence further decreases impedance. Seismic inversion was attempted to differentiate gas- and water-filled sandstones. Synthetic models were generated at the well locations using different fluid saturations to determine the effect of varying fluid type on the seismic signal. A supervised neural network, based on the synthetic models, was trained to look for a nonlinear relationship between impedance log and seismic traces. Neural-network validation was achieved by systematically hiding data from the network, rerunning the training session, and then observing predicted versus known relationships. The result of the inversion produced an improved quantitative calibration of the seismic amplitudes and captured the range of gas and water occurrences at the reservoir intervals.