Interpolation method based on pattern-feature correlation

Reasonable low-wavenumber initial models are essential for reducing the non-uniqueness of seismic inversion. A traditional approach to estimating the low-wavenumber models of elastic parameters is well-log interpolation. However, complex geological structures decrease the accuracy of this method. To overcome these challenges in building prior models, we propose an interpolation method based on pattern-feature correlation inspired by multiple-point geostatistics (MPG). In the proposed interpolation method, we scan a stacked seismic profile using a predefined data template to obtain a geological pattern around each node in the seismic profile. Each pattern is then converted into several filter-scores with the filters defined in the MPG algorithm of the filter-based simulation (FILTERSIM). We calculate the correlation coefficients of the filter-scores among different patterns for the various nodes and define them as the pattern feature correlations (PFCs). We construct the initial models from well-log data based on the weighted interpolation method, where the weighting factors are precisely determined by the PFCs. We build the initial models using the proposed method for both synthetic and field data to demonstrate its effectiveness. To verify the validity of the initial models, we apply them to Bayesian linearized inversion. The accuracy of the interpolation and inversion results verify the excellent performance of the proposed interpolation method. The proposed method provides a novel and convenient approach that combines seismic and well-log data, which contributes to both seismic exploration and geological modeling.