Seismic analysis, the process of identifying subsurface natural resources which are often worth billions of dollars, is expensive and time consuming. After raw sensor data has been processed to create a subsurface image, highly specialized analysts spend significant time and effort manually reviewing each image to identify and characterize the possible location of oil and gas. This is a lengthy and complex process requiring anywhere between 2-12 months and typically results in just a 30% success rate.

Challenges with
the traditional approach for seismic analysis

There have been several methods used to simplify and accelerate seismic analysis, typically focused on automating the more time consuming parts of the analysis process. However, these methods typically struggle to accurately automate analysis in areas where oil or gas are typically found, such as in settings where there are complex fault systems or depositional environments.

Challenges with the traditional approach for seismic analysis

Deep learning approaches to seismic analysis

Computer vision has the potential to greatly improve 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 tease out.

However, one of the biggest challenges is labeling this highly complex data before it can be interpreted by computer vision systems. Current approaches require up to 40% of the total volume of data to be labeled to train the model, with features that are complex, 3D in nature, and can only be labeled by highly-skilled hard-to-find experts.

SambaNova utilizes a proprietary approach that greatly simplifies the time consuming process of labeling data by reducing 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.

Deep learning approaches to seismic analysis

The Benefit of 3D data

Unsurprisingly, the best representation of real world 3D images is in 3D. By using large 3D inputs and deep networks enabled the SambaNova DataScale® platform, it is possible to leverage multiscale, 3D correlations in the image data not possible when analyzing 2D images. This results in substantially higher accuracy much faster.

SambaNova Advantage:Going 5123and beyond

Just like when analyzing 2D images, higher resolution 3D images enable better insights and more accurate analysis. However, the main limitation of 3D images is that their size increases exponentially as resolution grows. As a result, legacy GPU based platforms struggle to analyze images beyond 1283 due to the memory limitations.

3D images

Legacy GPU based platforms struggle to analyze images beyond 1283 due to the memory limitations. This results in clear degradation in quality of results, due to losing larger context and network depth.

3D Images

The higher memory capacity of the SambaNova DataScale platform enables using images 5123 and beyond. Natural correlations in the data can be retained resulting in higher quality of features with more richer details improving downstream analysis and accuracy of results.

loU and Accuracy Comparison

Seismic Data Terms of Use: The New Zealand government collects seismic and well data and releases it to the public after a data confidentiality period of a few years. The purpose of releasing these data to the public is to promote development of New Zealand’s petroleum and mineral resources. These data can be used by students, academics, and industry provided publications and presentations acknowledge New Zealand Petroleum and Minerals (NZPM) for providing data.


Label data source: AICrowd

Unlocking the multi-billion dollar opportunity trapped in seismic data

Unlocking the
multi-billion dollar opportunity
trapped in seismic data

The application of this state-of-the-art deep learning capability in analyzing 3D seismic images has huge potential for the oil and gas industry, where just a 1% improvement in discovery accuracy can increase profits by $100Ms.

Many oil and gas enterprises already have 100s of cubic miles of mapped seismic images, but lack the technology to unlock the necessary insights that can power this profitability.

With SambaNova true resolution computer vision, every oil and gas organization has the ability to analyze this volume of seismic data in record time using the most accurate and detailed 3D features available in the industry today to seize the multi-billion dollar opportunity trapped in their data.

Talk to an Expert

Learn how SambaNova can advance your AI initiatives to help you achieve your impossible.
Get Started