| Abstract Submission No. | ABS-2.4-0075 |
| Title of Abstract | Seismic facies identification using machine learning in Mahanadi offshore basin. |
| Authors | Suvonnita Saha*, Dr. Maheswar Ojha |
| Organisation | IIT ISM DHANBAD |
| Address | Research scholar hostel 1, CSIR NGRI, STADIUM HYDERABAD, Telangana, India Pincode: 500007 E-mail: suvonnitasaha@gmail.com |
| Country | India |
| Presentation | Oral |
| Abstract | The Mahanadi Offshore Basin, located along the eastern continental margin of India, represents a geologically complex environment where seismic facies interpretation is particularly challenging. Conventional manual approaches often struggle to delineate subtle stratigraphic features, leading to subjective and inconsistent outcomes. This study introduces a novel machine learning (ML) framework aimed at automating seismic facies classification to improve objectivity, reproducibility, and precision in subsurface interpretation. The methodology integrates multi-attribute seismic data¿including amplitude, frequency, phase, and coherence¿with well log information to train supervised classifiers such as Support Vector Machines (SVM) and Random Forests. High-quality 2D/3D seismic datasets from the basin are first enhanced through advanced pre-processing techniques, including band-pass filtering, FX deconvolution, and normalization. Seismic attributes are then extracted and optimized using Principal Component Analysis (PCA) for dimensionality reduction and feature selection. To ensure reliable ground-truth constraints, synthetic seismograms generated from well log data and Ricker wavelets are tied to seismic sections through time¿depth conversion and alignment. Model performance is assessed using metrics such as accuracy, F1-score, and confusion matrices, while predictive uncertainty is quantified via bootstrap resampling and Monte Carlo dropout. The anticipated outcomes include high-resolution seismic facies maps and uncertainty maps, which together provide enhanced insights into depositional settings, reservoir heterogeneity, and potential fluid distributions. This ML-driven approach offers a scalable, data-driven alternative to traditional interpretation techniques. By reducing subjectivity and improving predictive confidence, it holds the potential to significantly enhance exploration efficiency, mitigate risks, and optimize hydrocarbon exploration strategies in complex offshore basins like the Mahanadi. |
| Keywords | Seismic Facies, Machine Learning, Mahanadi Offshore Basin, Support Vector Machines, Random Forest, Seismic-Well Integration, Principal Component Analysis, Uncertainty Quantification, Hydrocarbon Exploration. |