A Scalable Framework for Flow Based Processing of Irregular N-Dimensional Seismic Volumes

SUMMARY

We present a novel framework for handling large, irregularly shaped and sampled, N-Dimensional seismic volumes that remains compatible with graph-flow-based seismic processing applications commonly used in our industry. Leveraging an abstract I/O framework, we demonstrate how to interconnect standard trace-wise and gather-wise processing steps with large volume-wise processing steps in a hardware-agnostic but scalable way.

This is crucial to allow algorithmic flexibility and quality control without compromising the HPC performance of the processing sequence. In fact, the abstracted nature of the framework decouples the HPC infrastructure from geophysical algorithms, allowing the same code to run performantly on a single small machine or supercomputers of various architectures.

We demonstrate use cases ranging from HPC crucial algorithms spanning data intense LS-RTM and LS-KDM down to simple filtering steps such as tapering irregular volumes.