Artificial Intelligence

SEAM AI

Pioneering AI & ML-Ready Benchmark Datasets – The First of Their Kind!

The SEAM AI project was initiated with the goal of advancing the state-of-the-art in artificial intelligence (AI) and machine learning (ML) applications in applied geoscience, particularly in seismic exploration, characterization, development, and management of underground fluid reservoirs. This includes petroleum (oil and natural gas) reservoirs and geological sequestration of hazardous materials such as large quantities of carbon dioxide to mitigate climate change.

The project is divided into three main components:

1. Identification, packaging, and provision of benchmark field datasets suitable for AI and ML applications.
2. Interpretation and labeling of selected subsets within those datasets.
3. Development of a seismic dataset of geophysically modeled reservoir properties from one of the benchmarked datasets.

Objectives and Benefits

• Identify and distribute suitable benchmark datasets for public use and assess AI applications on these data.
• Advance AI and ML applications in seismic exploration and reservoir management.
• Perform synthetic modeling to differentiate near-surface and total model responses of previous SEAM Earth models.
• Enable cloud-based collaboration for AI advancements in geophysics.
• Establish global standards for data exchange, cooperative projects, and research communication

Licensing SEAM AI Provides Organizations With:

• Access to AI and ML-ready labeled datasets
• Data covering multiple geological complexities and facies worldwide
• Access to subsets of previous SEAM models, including Barrett, Arid, Foothills, and Life of Field projects

Technical Specifications

The datasets were acquired from vendors, quality-controlled, and cataloged to ensure consistency across data quality, coordinate systems, phase, and amplitude. These datasets span various marine depositional environments, geographically diverse basins, and include onshore and offshore data from shallow-water to deep-water settings. Where available, datasets include:

• Depth and/or time-migrated 3D seismic reflection data.
• Corresponding interval or RMS velocity data.
• Digital well data from within the 3D seismic survey area.
• Supporting remote sensing datasets.
• Cultural data relevant to the study areas.
• Pertinent background research materials.

All selected areas of interest are covered by 3D and/or 4D seismic surveys available in the public domain. These datasets underwent an interpretation phase to identify regional seismic facies, enabling developers to create algorithms addressing fundamental complexities encountered in seismic interpretation and exploration.

The selected areas, either entire volumes or subsets, aim to create seismic attribute volumes with values assigned to differentiate seismic facies. Final input volumes were chosen for regional seismic facies mapping across the following datasets:

Additional SEAM AI Products

In addition to labeled seismic facies volumes, SEAM AI has generated prepared data for synthetic, noise, and interpolation modeling using existing SEAM Earth models:

• Arid (Desert environments, including karsts)
• Foothills (Mountainous regions)
• Barrett (Unconventional reservoirs)
• Life of Field Clastic (4D reservoir modeling)
• Life of Field Carbonate (4D reservoir modeling)

These additional products were developed through three key tasks:

1. Synthetic seismic modeling to separate near-surface and subsurface seismic responses.
2. Event identification and travel-time picking for synthetic data using ray tracing.
3. Preparation of SEAM datasets for noise and interpolation

Deliverables

All deliverables produced by SEAM AI are in SEG-Y format. Labeled volumes contain facies codes, with survey geometries applied to enable direct overlay and comparison with original seismic volumes. These facies volumes are ready for direct integration into machine learning algorithms as benchmark datasets for geophysical applications.

Participants