SEISMIC LITHOLOGIC MODELLING

Abstract

Seismic data constitute a noise-contaminated bandlimited representation of lithologic information. Therefore, attempts to recover lithologic information, such as velocity and density of thin layers, from seismic data cannot give a unique or verifiably correct result. Nevertheless, an interpreter must produce detailed geologic models on the basis of such data.

Our interactive modelling procedure quantitatively refines the model that the interpreter derives from available geologic, borehole, and seismic data. The procedure starts from the initial model and perturbs its parameters (velocity, density, and layer thickness) to find the least-squares best fit between the stacked field data and the synthetic section derived from the model. The layer parameters are explicitly varied only at selected locations along the profile and are linearly interpolated to intermediate positions. This approach is based on an assumption of smooth lateral change for parameters, and enables the process to see through much of the noise.

Synthetic data tests demonstrate that, even for significantly different initial models, the procedure can find solutions that are close to the correct model. Moreover, the procedure is stable in the presence of noise and not severely dependent on the shape of the bandlimited seismic signal or on details of the initial hypothesis.

Application of the method to stacked data from the Denver Basin provides evidence that an interpreter's model can be refined so as to better match the seismic data while honouring available non-seismic data as well. A low-velocity anomaly in the final solution is well correlated with a known hydrocarbon play; the anomaly disappears near the dry well. Application of the method to stacked data from widely different data environments further demonstrates the value of the technique for refining thin-layer interpretations, thus providing, in effect, a geology-based deconvolution of seismic data.