This study documents reservoir delineation and gas thickness prediction for a thin, poorly consolidated, porous deltaic sand in a Miocene basin. A grid of 4 seismic lines over the field was calibrated with 4 wells. Nine seismic attributes (amplitudes, areas, and thicknesses) were measured for statistical analysis. The choice of measurements was guided by a simple physical model of the seismic response of a thin, bright reservoir of fixed acoustic impedance embedded in thick shales. A learning set, composed of 420 seismic traces where the pore fluid was assumed known, was chosen based on information from predevelopment wells. A multivariate discriminant function, based on the 9 attributes, correctly differentiated gas from brine for 98% of the learning set. Application of his function to the seismic grid produced a map of the gas-water contact consistent with the contact located by development drilling.

Multivariate regression analysis was used to relate gas thickness to seismic attributes for 14 traces adjacent to wells. A 5-variable prediction equation was the most accurate model at well locations, but the predictions were overly sensitive to noise away from the wells. Use of a single amplitude measurement reduced this sensitivity. The single-variable model, in conjunction with the multivariate gas or brine discriminator, resulted in a gas reserve estimate close to the post-development estimate. This study provides a rapid method for pore-fluid discrimination and net-pay prediction in the production setting. The empirical nature of the resulting statistical functions limits their application to the specific field for which they are derived. However, given a few wells and a reasonab e geologic model, predictors for new fields can be developed easily.

End_of_Article - Last_Page 285------------