| Abstract |
This study introduces an innovative approach to 3D seismic facies interpretation by using an advanced deep learning model that enhances the conventional long short-term memory unit with convolution operations. This modification allows for the effective capture of spatial and temporal dependencies, addressing the limitations of manual 2D interpretation methods that are typically labor-intensive and inaccurate. Through comparative experiments, we demonstrate the superiority of our proposed model over traditional methods that focus solely on spatial dependencies. The findings underscore the importance of incorporating both spatial and temporal dependencies for more accurate boundary detection and lithostratigraphic unit classification. The proposed model, in particular, outperforms traditional convolutional neural networks by providing more accurate predictions and effectively identifying lithostratigraphic boundaries. Additionally, our study highlights the potential of this deep learning approach to serve as an automatic seismic interpretation method that requires just a few labeled profiles, making it a valuable tool for preliminary subsurface characterizations and early-stage geological explorations. Future work may explore refining predictions with post-processing strategies to improve the accuracy of seismic facies interpretation further. |
, Sumit Verma 
