Deep Learning Workflow for Velocity Model Building

This technical report presents a deep learning-based workflow aimed at accelerating the process of Velocity Model Building (VMB) in seismic imaging, particularly in frontier exploration settings. The report outlines the challenges associated with traditional VMB methods, which often rely on poorly constrained initial models and extensive manual intervention. The proposed workflow integrates advanced machine learning techniques, including Convolutional Neural Networks (CNNs) and Fourier Neural Operators (FNOs), to automate key steps in the VMB process. It focuses on three components: estimating a robust initial macro-velocity model, automating the picking of complex water-bottom horizons, and replacing conventional tomography methods. The methodology is validated using field data from various sedimentary basins, including the Agung area in Bali. The results demonstrate significant improvements in model accuracy and turnaround time, effectively streamlining the VMB workflow and enhancing the quality of initial models for Full Waveform Inversion (FWI).