Waveform Gather Inversion and Attribute-Guided Interpolation: A Two-Step Approach to Log Prediction

By

 August Lau¹, Alfonso Gonzalez², Subhashis Mallick², Diana Gillespie²

(1) Apache Corporation, Houston, TX (2) WesternGeco, Houston, TX

 We present a two-step approach to predict log information. In the first step nonlinear waveform gather inversion is used to estimate VP, VS, and density from the full seismic gathers. This estimation is useful in areas with little or no well control, as well as in mature basins where log information might be available but might not be complete. Waveform gather inversion is computationally intensive, and therefore, difficult to apply to every gather in a seismic volume. In the second step, seismic attributes are used to guide the interpolation of the predicted logs for the entire seismic volume. Interpolation of log properties can be done in several ways. A traditional way is to use hybrid inversion, elastic impedance inversion, or poststack amplitude inversion. These methodologies depend exclusively on amplitude information and use no other attributes. This traditional inversion is also strongly dependent on the interpretation of horizons. The interpretation in certain areas could be challenging, as in the following types of geologic settings: carbonate buildup, channel complex, crossing fault, angular unconformity, or turbidite. Furthermore, certain poststack inversions might be too sensitive to the interpretation, which is not always desirable. Neural network interpolation does not require horizon interpretation. Our two-step approach overcomes the high cost of waveform gather inversion, and results in a parallelized workflow where the inversion and neural network interpolation are done independently of the structural interpretation of the seismic data. This overcomes the bottleneck of a linear workflow of processing, interpretation, and inversion.