High resolution 3D seismic analysis
The multi billion dollar opportunity trapped in seismic images
Accelerating complex seismic analysis
Seismic analysis is a complex, 12 month process that has a 70% failure rate in accurately predicting discoveries. The process involves identifying subsurface natural resources, such as oil or gas deposits, which represent billions in potential profit for the oil and gas organizations.
To predict the location of these accumulations, highly trained specialists need to analyze hundreds of km3 of extremely detailed geospatial image data. The accuracy of these predictions is critical, as an incorrect prediction not only represents billions in missed opportunity but a massive wasted capital investment of $100M or more.
In order to maximize the success of these predictions, oil and gas organizations invest months or years in highly complex and time consuming analysis that not only creates a massive bottleneck in the full geoscience decision workflow, but typically results in a 70% failure rate in correctly predicting the location of these valuable deposits.
TRUE RESOLUTION COMPUTER VISION
Improving accuracy of discovery predictions
Deep learning computer vision models have the potential to greatly improve and accelerate the analysis of seismic data by capturing even higher dimensional information, such as amplitude versus offset, as part of the patterns that experts need to identify. This enables reducing the seismic analysis process from nine weeks to one week.
Despite years of digital innovation, seismic interpretation remains as a cumbersome and largely manual activity, with typical projects spanning 2-12 months and remains a core exploration, production, and CCS geoscience bottleneck.
SambaNova Systems’ deep learning computer vision models can greatly improve and accelerate the analysis of seismic data by capturing signal in 3 dimensions and beyond as part of the patterns that experts need to identify.
Deep learning computer vision models can accelerate time to analysis by 13% or more. This improved analysis speed enables oil and gas organizations to analyze and make 13% more predictions, worth an estimated average of $2.7B annually.
TRAINING COMPUTER VISION MODELS
Accelerated training of seismic computer vision models
Training computer vision models for seismic analysis requires specialized data labeling of image data sets hundreds of km3 in size. SambaNova utilizes a proprietary approach that reduces the labeled data required to train a model by 97.6%, while resulting in more detailed 3D features that can greatly improve the accuracy of the analysis.
The labeling process for these computer vision models is highly complex, requiring significant manual effort from subject matter experts that can recognize and label the complex 3D features critical to seismic analysis and prediction.
SambaNova utilizes a proprietary model training approach that reduces the labeled data required to train a model by 97.6%, while simultaneously resulting in more detailed 3D features that can greatly improve the accuracy of the analysis.
This reduction in the complex and highly specialized data labeling requirement for seismic computer vision models makes computer vision practical for energy organizations, enabling them to improve and accelerate their analysis at scale
HIGH RESOLUTION 3D SEISMIC IMAGES
High resolution seismic images: 5123 and beyond
While 2D networks can be useful for seismic analysis, they can lead to artifacts and complicated post-processing issues that impact usability of the prediction. Large 3D networks overcome these limitations by leveraging multi-scale, 3D correlations to identify the most relevant features in the data to improve prediction quality.
The 3D image resolution that can be processed with GPU-based AI infrastructure is extremely limited, leading to tiling or lowering the resolution of images, which both increases implementation complexity and significantly impacts accuracy.
The higher memory capacity of the SambaNova DataScale® platform enables analyzing images at 5123 resolution and beyond, identifying natural correlations and higher quality of features, improving downstream analysis and accuracy of results.
The higher quality of features from 5123 resolution images enables seismic analysts to make more accurate predictions in less time, resulting in potential for billions of dollars from the discovery of additional resource deposits.
Label data source: AICrowd
Request a meeting or demo to learn more
Learn how SambaNova can help your organization leverage AI to competitive advantage.
Stay on Top of AI: trends, information and company news delivered to you.