Deep Learning Can Now Interpret Seismic the Way Experts Do

The SubsurfaceAI custom deep learning application for seismic allows experts to annotate data, identify sequences and, in this example, define a fault complex. This forms the basis of a workflow that allows a seismic expert to apply deep learning to ‘interpret the way experts do,’ creating bespoke models for seismic interpretation. 

Author: Ben Lasscock, Ph.D.

2 Years Seems So Long Ago   

During the keynotes and panels of the SEG 2018 annual meeting, there were general calls to action for managers and business leaders to move faster on the creation and application of digital technologies. The intent was to instill a sense of urgency for transforming the way they work and to aspire to new possibilities for innovation. A number of discussions pointed out the directly applicable digital work in other industries – life sciences, for example, where visualization and image analysis using AI/machine learning are well ahead of the energy industry.

SEG 2019 had similar messages, and there was evidence of progress. At the event, Enthought invited attendees to visit the booth and train a deep learning model for automating seismic interpretation. This was by no means cutting edge when seen within a larger family of science-driven businesses. However, in the energy industry, it was highly innovative.

However, urgency in an organization is not created without, well, urgency. And then came 2020.

Enter Black Swans    

The double black swan events, or perhaps the bevy of events, for the oil industry in 2020 have provided significant urgency to the adoption of digital technology, which had been missing, despite the price collapse of 2014.

McKinsey & Company conducted a survey of executives on rates of digital adoption. The results, published in a 5 October 2020 article, note that companies have accelerated implementations of digital technologies by 10 to 40 times over plan, particularly in the area of remote working and collaboration.

This quote from the McKinsey article says it all:

For many of these changes, respondents say, their companies acted 20 to 25 times faster than expected. In the case of remote working, respondents actually say their companies moved 40 times more quickly than they thought possible before the pandemic. Before then, respondents say it would have taken more than a year to implement the level of remote working that took place during the crisis. In actuality, it took an average of 11 days to implement a workable solution, and nearly all of the companies have stood up workable solutions within a few months.

When respondents were asked why their organizations didn’t implement these changes before the crisis, just over half say that they weren’t a top business priority. 

Hello, urgency. Meet Seismic.

Looking Ahead at SEG20 

This same urgency that McKinsey highlights for the adoption of digital technologies is now accelerating their acceptance in the oil industry. Adoption in areas centered on data is relatively straightforward as compared to where there are changes implied to ‘digitally advance’ physical assets.

What does this all mean for seismic processors, interpreters and their managers? First, they will now be working business problems with a clear sense of urgency. Cost pressure, asset transactions and layoffs are strong indicators of seismic urgency. However, will there still be a budget for developing new ideas without an expectation for an immediate return on investment?

Starting with the end in mind, AI/machine learning tools in the geoscience domain will ultimately have to be flexible and customizable without requiring the user to have a degree in data science or statistics. This is the norm on the web right now; we don’t need to understand a search engine to use the internet effectively or how to work with maps on our smartphones to find the best route home. We just do.

Instead of starting with an academic mindset, imagine if we can start by getting intuitive deep learning applications directly into the hands of a subsurface team working projects. These tools will be good enough to overcome the initial cost of their uptake and deliver immediate value. Improving capabilities over time of this new generation of deep learning tools will expand the areas of geoscience where new business problems can be tackled.

Enthought has been working to create applications that start with the earth scientist in mind. In seismic, specialist AI/machine learning algorithms within web applications have been developed to allow an interpreter to identify some feature in a large seismic volume or label several lines for facies and have the results waiting when back from lunch. This same approach has been taken for core-CT scans of thin sections, where large amounts of high-resolution image data have lain dormant for years, just missing the power of deep learning.

Come ‘chat’ with us at the Virtual SEG 2020, or visit us online to learn more about the urgency that ‘industry urgency’ created in Enthought. Find out how you and your team can start crafting deep learning applications and get them innovating.

About the Author

Ben Lasscock holds a Ph.D. and a B.Sc. in theoretical physics as well as a B.Sc. in physics and theoretical physics from the University of Adelaide. Before coming to geoscience, Ben worked as a portfolio manager at a large hedge fund in Australia. He has publications in the areas of high energy physics, Bayesian time series analysis and geophysics.

 

Share this article:

Related Content

3 Trends for Scientists To Watch in 2023

As a company that delivers Digital Transformation for Science, part of our job at Enthought is to understand the trends that will affect how our…

Read More

Retuning the Heavens: Machine Learning and Ancient Astronomy

What can we learn about machine learning from ancient astronomy? When thinking about Machine Learning it is easy to be model-centric and get caught up…

Read More

Extracting Target Labels from Deep Learning Classification Models

In the blog post Configuring a Neural Network Output Layer we highlighted how to correctly set up an output layer for deep learning models. Here,…

Read More

Exploring Python Objects

Introduction When we teach our foundational Python class, one of the things we do is make sure that our students know how to explore Python…

Read More

Choosing the Right Number of Clusters

Introduction When I first started my machine learning journey, K-means clustering was one of the first algorithms I was introduced to – and it is…

Read More

Prospecting for Data on the Web

Introduction At Enthought we teach a lot of scientists and engineers about using Python and the ecosystem of scientific Python packages for processing, analyzing, and…

Read More

Configuring a Neural Network Output Layer

Introduction If you have used TensorFlow before, you know how easy it is to create a simple neural network model using the Keras API. Just…

Read More

No Zero Padding with strftime()

One of the best features of Python is that it is platform independent. You can write code on Linux, Windows, and MacOS and it works…

Read More

Got Data?

Introduction So, you have data and want to get started with machine learning. You’ve heard that machine learning will help you make sense of that…

Read More

Sorting Out .sort() and sorted()

Sorting Out .sort() and sorted() Sometimes sorting a Python list can make it mysteriously disappear.  This happens even to experienced Python programmers who use .sort()…

Read More

Join Our Mailing List!

Sign up below to receive email updates including the latest news, insights, and case studies from our team.