Leveraging GPU-Based High-Performance Computing for Enhanced Wavefield Traveltime Inversion Algorithms

Abstract Full Waveform Inversion (FWI) has become a widely accepted method for accurate velocity estimation in seismic imaging and reservoir modeling, but it still faces challenges, particularly in building a suitable starting model and managing computational resources. To overcome these limitations, we have developed the Wavefield Traveltime Inversion (WTI) algorithm, which provides a reliable initial model for FWI. We have optimized the WTI algorithm for efficiency on a Graphics Processing Unit (GPU) system by rewriting its functions in the CUDA programming language, employing a GPU-based lossy compression library to reduce memory requirements, and utilizing multiple CUDA streams for enhanced performance. Additionally, we have modified the task distribution system to better suit the WTI program, ensuring that computational resources are utilized efficiently. The resulting GPU-based WTI algorithm has been applied to a 3D land dataset, demonstrating its capabilities in providing accurate and detailed velocity models, and highlighting its potential as a valuable tool for seismic imaging and reservoir modeling, particularly when used in conjunction with FWI.