Application of a density-based spatial clustering algorithm for fully automatic picking of surface-wave dispersion curves

Surface waves can be used to enhance the characterization of the shallow subsurface in desert environments. A high-resolution shear-wave velocity model is typically obtained by inverting dispersion curves, which correspond to different propagation modes of the surface waves. A common approach to estimate the dispersion curves is to manually pick the magnitude maxima from the frequency-phase velocity spectra of the seismic data. This approach is inefficient, time consuming, highly subjective, and not feasible for large surveys. Automatic picking of dispersion curves has become a topic of interest recently in the oil and gas research community, where many of the developed algorithms were inherited from the fields of image processing and machine learning. By exploring in the area of unsupervised learning, we recently derived an algorithm and workflow for fully automatic picking of surface-wave dispersion curves by employing a density-based spatial clustering technique. Our approach has been tested on the SEG Advanced Modeling Corporation Arid model synthetic data set and a field data set acquired in a desert environment. The results of the synthetic tests show that the estimated dispersion curves match the true dispersion curves with high accuracy, and they can be inverted for shear-wave velocities, successfully recovering the shallow near-surface features. The application of the method to field data provides high-resolution geology-consistent shear-wave velocity information that can be converted into a compressional-wave velocity model in agreement with uphole observations.