Machine Learning Framework for Automated Seismic Facies Classification in the Niger Delta Deepwater

This study develops and validates an original deep learning framework for automated seismic facies classification in the Niger Delta deepwater, contributing a methodology that significantly improves the speed and objectivity of seismic interpretation for turbidite reservoir characterisation. Manual seismic facies interpretation is subjective, time-consuming, and difficult to apply consistently across large 3D seismic volumes. Machine learning provides an alternative that can extract non-linear feature combinations from seismic attributes beyond the capacity of conventional crossplot analysis. This study develops a hybrid convolutional recurrent neural network architecture specifically designed for volumetric seismic facies classification that exploits both spatial pattern recognition and temporal stratigraphic sequence information. The network is trained on a labelled training dataset comprising 1.2 million 3D seismic voxels from 12 deepwater fields with well-calibrated facies interpretations, using data augmentation to address class imbalance between thin and thick facies. Transfer learning adapts the pre-trained network to new fields using limited labelled data. Interpretability analysis using gradient-weighted class activation mapping identifies the seismic attributes most influential in each classification decision. Findings demonstrate that the hybrid architecture achieves overall facies classification accuracy of 88 percent on held-out test data, compared to 72 percent for conventional attribute-based neural networks and 65 percent for single-attribute methods. Thin-bed turbidite facies (below seismic tuning) are classified correctly in 68 percent of cases, representing a significant advance over conventional methods. Field applications in three previously interpreted deepwater blocks show high consistency with expert interpretations while reducing interpretation time by 85 percent. The study contributes an original deep learning seismic interpretation tool for the Niger Delta.

Keywords: deep learning, seismic facies classification, Niger Delta, convolutional neural network, turbidite reservoir.