AI Applied Geoscience Hackathon results released

Seven teams of geoscientists and data scientists collaborated and competed 22 March–9 April

SEG Advanced Modeling project (SEAM) OpenACC Organization, and NVIDIA co-hosted the SEAM AI Applied Geoscience GPU Hackathon from 22 March–9 April. Seven teams of geoscientists and data scientists collaborated and competed in developing solutions for two hard challenges in the use of machine learning and artificial intelligence to understand large seismic datasets. The first challenge involved a six-way “facies classification” of pixels in a 3D image of earth’s interior produced from marine seismic data collected during petroleum exploration off the northwest coast of New Zealand. The second challenge involved separation of signal from near-surface geologic noise in land seismic data.

The facies classification challenge, using the public-domain Parihaka dataset available from the New Zealand Government, is a small example of the full problem of seismic interpretation, in which an image of earth’s subsurface is subdivided into geologic facies (rock types) corresponding to different environments present at earth’s surface when the now-buried sediments were deposited. Identification of different sedimentary environments in geologic formations can provide important clues to the location of underground petroleum or groundwater reservoirs or to locations where industrial wastes can be safely sequestered for long times. The Parihaka challenge had been posed before as part of a workshop hosted by the Society of Exploration Geophysicists (SEG) at its 90th Annual Meeting in October 2020 and was also part of a public competition hosted by SEAM on the AIcrowd machine learning challenge platform. The goal of the GPU Hackathon was to develop more refined classification techniques using scoring metrics better informed by geologic reasoning. For ground truth, the facies challenge used a classification of the Parihaka image by an expert geologist, which the energy company Chevron had donated to the SEG for its 90th Anniversary Meeting.

The signal recovery challenge used a realistic synthetic dataset simulating land seismic exploration over unconventional shale reservoirs. The Barrett Unconventional Model, which generated the synthetic data, was produced in an earlier SEAM project involving 22 major oil and oilfield service companies. The model was simulated in a novel way allowing separation of the synthetic data into signals coming from the deep subsurface, where the shale reservoirs are buried, and signals coming from the first 100 meters of the ground. This near-surface region can generate complex seismic responses such as surface waves, obscuring what lies beneath–like looking through a mud-covered window to see what’s outside. The signal recovery challenge was the first of its kind in the public domain.

Technical experts from Chevron, Aramco, Shell, Xrathus, and NVIDIA helped to mentor the participants, providing insights on geoscience, machine-learning, and high-performance computing. After training and collaborating for two weeks on a GPU cluster hosted by NVIDIA, participating teams submitted their best answers to one or both challenges. Three teams shared top honors in the seismic facies challenge: a team of data scientists from the software company MathWorks, a team of geoscientists and data scientists from French energy company Total, and a team of young geoscience professionals from Nigeria. For the signal recovery challenge, the Total team scored the highest with two other teams scoring well: a team of reservoir engineers from College Station (Texas) and a diverse group of geoscientists, software engineers, and data scientists from the Brazilian energy company Petrobras.

SEAM and NVIDIA plan to continue collaborating on machine-learning challenges in applied geoscience requiring high-performance computing during the remaining 18 months of the SEAM AI project.

SEAM, a subsidiary of the SEG, organizes collaborations among industry, government, and academia to address major industry subsurface challenges. The OpenACC Organization is dedicated to helping the research and developer community advance science by expanding their accelerated and parallel computing skills. NVIDIA is the pioneer of accelerated computing — a supercharged form of computing at the intersection of computer graphics, high-performance computing, and AI.