Abstract

We have applied instantaneous seismic attributes to a stacked P-wave reflected seismic section in the Tenerife field located in the Middle Magdalena Valley Basin in Colombia to estimate the volume of clay Vclay and the density ρ at seismic scale. The well logs and the seismic attributes associated to the seismic trace closer to one of the available wells (Tenerife-2) is the information used to train some multilayered artificial neural networks (ANN). We perform data analysis via the gamma test, a mathematically nonparametric nonlinear smooth modeling tool, to choose the best input combination of seismic attributes to train ANNs to estimate Vclay and ρ. Once the ANNs are trained, they are applied to predict these parameters along the seismic line. From the continuous estimations of Vclay, we distinguish two facies: sands for Vclay<0.5 and shales when Vclay0.5. These estimations confirm the production of the Mugrosa C-Sands zone, and we draw the brown shale that correlates with the high-amplitude attributes and the yellow sand that correlates with the low-amplitude attributes. Using the well-log information for VP and the facies classification (also in the well log), two cubic polynomials that depend on time (or depth) are obtained, one for sands and the other for shales, to fit the VP. These two cubic polynomials and the facies classification obtained from the Vclay at the seismic scale enable us to estimate VP at the seismic scale. To validate the 2D VP and ρ predicted data, a forward-modeling software (the Kennett reflectivity algorithm) is used. This model calculates synthetic seismograms that are compared with the real seismograms. This comparison indicates a small misfit that suggests that the VP and ρ images are representing the reservoir description characteristics and the ANN method is accurate to map these parameters.

You do not currently have access to this article.