Deep Learning-based Low Frequency Extrapolation: Its implication in 2D Full Waveform Imaging for Marine Seismic Data in the Sadewa Field, Indonesia

Seismic data with a low-frequency content is crucial for full waveform imaging (FWI) as it can improve the resolution of subsurface features and offer insight into the underlying geological characteristics. A lack of low-frequency content may cause cycle-skipping, which can distort the results of the inversion process. Low frequency content in seismic data is usually estimated using seismic processing-based deghosting techniques. In this study, an attempt is made to reconstruct the low-frequency content using an artificial intelligence approach through a deep learning algorithm. A convolutional neural network (CNN) approach was used to automatically extrapolate the low-frequency content of the band-limited common-shot-gather data, without the need for preprocessing steps. The model was first tested and validated with synthetic data. The optimised model was applied to the Sadewa field, Indonesia, and the obtained low-frequency extrapolated data were used as input for the FWI process. The results show that the proposed algorithm can reliably extrapolate the low-frequency content of the field data with minimal errors and exhibits good agreement with the deghosting results. The FWI results also demonstrate that our proposed method can be a reliable and efficient substitute for determining the low-frequency component of seismic reflection data.