We have trained deep convolutional neural networks (DCNs) to accelerate the computation of seismic attributes by an order of magnitude. These results are enabled by overcoming the prohibitive memory requirements typical of 3D DCNs for segmentation and regression by implementing a novel, memory-efficient 3D-to-2D convolutional architecture and by including tens of thousands of synthetically generated labeled examples to enhance DCN training. Including diverse synthetic labeled seismic in training helps the network generalize enabling it to accurately predict seismic attribute values on field-acquired seismic surveys. Once trained, our DCN tool generates attributes with no input parameters and no additional user guidance. The DCN attribute computations are virtually indistinguishable from conventionally computed attributes while computing up to 100 times faster.

The geophysical world has been recording seismic waves for over a century now, with the seismograph seeing first utilization towards exploration of oil and gas in the early 1900s. We started shooting 3D seismic data half a century later, and since then both our acquisition methods and how we reconstruct waves propagating through the earth with hyper-computing capabilities has evolved tremendously. Pushing the envelope of what we can image, in particular, is a major tranche of changes in seismic acquisition that started roughly a decade ago. Some of these recent developments include broader-bandwidth seismic acquisitions, particularly emphasizing low frequencies for both land and marine, and changes to sensors, sensor layouts and patterns used for shooting. These new acquisitions have refocused our emphasis on fundamentals in seismic processing, significantly advancing our ability to see the subsurface. Some of the ideas in seismic processing formulated for these acquisitions, not surprisingly, are not exclusively applicable only for modern acquisitions. Combining some of these newer approaches with pioneering ideas previous generations of geophysicists mastered, has allowed a fresh take on how large archives of legacy seismic, sitting-on-shelves, can be improved to provide fresh insights towards exploration. For the Middle East and North Africa region, where we often deal with ‘difficult’ seismic typically characterized by extremely high noise content, this re-look at older data has resulted in an evolution of workflows for vintage seismic data conditioning, leading to higher quality datasets that increase confidence and reduce uncertainty for the plays, leads and prospects we pursue. The data conditioning workflow involves a number of steps, and are mostly applied post-migration and often post-stack. These are applicable across a spectrum of data types from different sources and geological settings. Typically, these workflows are fine-tuned for each seismic dataset in a matter of days or often ‘on-the-fly’, with implementation on the next generation interpretation platform in Shell, providing extremely rapid turnaround, resulting in dramatic uplift in image quality in many cases. These have demonstrably impacted decision-making in exploration and production, providing the ability to, quite simply, see with significant more clarity what we could not see before. We share examples of utilization of these workflows, contributing towards a number of projects, including an extensive joint Abu Dhabi National Oil Company (ADNOC) and Shell effort to rejuvenate the Exploration Portfolio of Abu Dhabi, working with a country-wide database of multi-vintage onshore and offshore datasets constituting approximately 2500 2D seismic lines and more than 50 3D seismic volumes. Our workflows are grounded in past experience, yet leverage latest signal-processing innovations. They provide a step-change in our ability to rapidly investigate and interpret large volumes of challenging seismic data efficiently, in addition to enabling visualization of geological features indistinguishable on original seismic.

Traditional seismic processing workflows (SPW) are expensive, requiring over a year of human and computational effort. Deep learning (DL) based data-driven seismic workflows (DSPW) hold the potential to reduce these timelines to a few minutes. Raw seismic data (terabytes) and required subsurface prediction (gigabytes) are enormous. This large-scale, spatially irregular time-series data poses seismic data ingestion (SDI) as an unconventional yet fundamental problem in DSPW. Current DL research is limited to small-scale simplified synthetic datasets as they treat seismic data like images and process them with convolution networks. Real seismic data, however, is at least 5D. Applying 5D convolutions to this scale is computationally prohibitive. Moreover, raw seismic data is highly unstructured and hence inherently non-image like. We propose a fundamental shift to move away from convolutions and introduce SESDI: Set Embedding based SDI approach. SESDI first breaks down the mammoth task of large-scale prediction into an efficient compact auxiliary task. SESDI gracefully incorporates irregularities in data with its novel model architecture. We believe SESDI is the first successful demonstration of end-to-end learning on real seismic data. SESDI achieves SSIM of over 0.8 on velocity inversion task on real proprietary data from the Gulf of Mexico and outperforms the state-of-the-art U-Net model on synthetic datasets.