We have developed nonlinear beamforming (NLBF), a method for enhancing modern 3D prestack seismic data acquired onshore with small field arrays or single sensors in which weak reflected signals are buried beneath the strong scattered noise induced by a complex near surface. The method is based on the ideas of multidimensional stacking techniques, such as the common-reflection-surface stack and multifocusing, but it is designed specifically to improve the prestack signal-to-noise ratio of modern 3D land seismic data. Essentially, NLBF searches for coherent local events in the prestack data and then performs beamforming along the estimated surfaces. Comparing different gathers that can be extracted from modern 3D data acquired with orthogonal acquisition geometries, we determine that the cross-spread domain (CSD) is typically the most convenient and efficient. Conventional noise removal applied to modern data from small arrays or single sensors does not adequately reveal the underlying reflection signal. Instead, NLBF supplements these conventional tools and performs final aggregation of weak and still broken reflection signals, where the strength is controlled by the summation aperture. We have developed the details of the NLBF algorithm in CSD and determined the capabilities of the method on real 3D land data with the focus on enhancing reflections and early arrivals. We expect NLBF to help streamline seismic processing of modern high-channel-count and single-sensor data, leading to improved images as well as better prestack data for estimation of reservoir properties.
We have developed a novel approach for seismic modeling combining conventional finite differences with deep neural networks. The method includes the following steps. First, a training data set composed of a small number of common-shot gathers is generated. The data set is computed using a finite-difference scheme with fine spatial and temporal discretization. Second, the entire set of common-shot seismograms is generated using an inaccurate numerical method, such as a finite-difference scheme on a coarse mesh. Third, the numerical dispersion mitigation neural network is trained and applied to the entire data set to suppress the numerical dispersion. We have tested the approach on two 2D models, illustrating a significant acceleration of seismic modeling.
8 September 2020 in Geophysical Journal International
SUMMARY: We propose an advanced version of non-linear beamforming assisted by artificial intelligence (NLBF-AI) that includes additional steps of encoding and interpolating of wavefront attributes using inpainting with deep neural network (DNN). Inpainting can efficiently and accurately fill the holes in waveform attributes caused by acquisition geometry gaps and data quality issues. Inpainting with DNN delivers excellent quality of interpolation with the negligible computational effort and performs particularly well for a challenging case of irregular holes where other interpolation methods struggle. Since conventional brute-force attribute estimation is very costly, we can further intentionally create additional holes or masks to restrict expensive conventional estimation to a smaller subvolume and obtain missing attributes with cost-effective inpainting. Using a marine seismic data set with ocean bottom nodes, we show that inpainting can reliably recover wavefront attributes even with masked areas reaching 50–75 per cent. We validate the quality of the results by comparing attributes and enhanced data from NLBF-AI and conventional NLBF using full-density data without decimation.