Benjamin Consolvo

Salt interpretation on seismic data has historically been a very manual process, requiring weeks or even months to complete on one 3D seismic survey. The accuracy of the interpreted salt boundary is critical for sub-salt imaging and subsequent drilling for oil and gas. The nature of the salt problem can be reduced to a binary classification problem that is well suited to modern machine learning (ML) algorithms: each location on an image either contains salt or sediment. Seismic surveys are… Show more

Salt interpretation on seismic data has historically been a very manual process, requiring weeks or even months to complete on one 3D seismic survey. The accuracy of the interpreted salt boundary is critical for sub-salt imaging and subsequent drilling for oil and gas. The nature of the salt problem can be reduced to a binary classification problem that is well suited to modern machine learning (ML) algorithms: each location on an image either contains salt or sediment. Seismic surveys are collected and processed in different ways, which poses a challenge to traditional ML methods that rely on statistical similarity between training data and prediction data, especially where limited training data are available. We propose to use a supervised ML approach that treats each seismic survey independently. In particular, we show that an adaptive U-Net approach yields accurate salt bodies in minutes rather than weeks and requires minimal training data. Show less

Microseismic monitoring is a crucial element to understanding hydraulic fracturing operations prior to oil and gas production. One of the more tedious quality control (QC) measures that must often be performed following a microseismic processing workflow is a visual inspection of seismic data to determine whether the data contain microseismic events or only noise. The manual nature of these inspections can take many weeks, sometimes over a month, to perform for one geophysicist. Automated… Show more

Microseismic monitoring is a crucial element to understanding hydraulic fracturing operations prior to oil and gas production. One of the more tedious quality control (QC) measures that must often be performed following a microseismic processing workflow is a visual inspection of seismic data to determine whether the data contain microseismic events or only noise. The manual nature of these inspections can take many weeks, sometimes over a month, to perform for one geophysicist. Automated approaches usually use a short-term-average long-termaverage (STA/LTA) ratio, but end up picking false positives on noisy data. We propose using a supervised deep learning algorithm, a convolutional neural network (CNN), to automatically classify microseismic events from noise. Using our deep learning approach, we show that the time for QC can be reduced from weeks to hours with high accuracy. Show less

Interpreting salt bodies in the subsurface is a challenging manual task that can take weeks to complete. Obtaining accurate picks of salt is very important, because errors in the placement of salt can result in severe degradation of the seismic image. To meet the challenges of speeding up imaging workflows and retaining accurate salt picks, we evaluate three deep learning approaches: a 2D window-based convolutional neural network, a 3D window-based convolutional neural network, and finally a 2D… Show more

Interpreting salt bodies in the subsurface is a challenging manual task that can take weeks to complete. Obtaining accurate picks of salt is very important, because errors in the placement of salt can result in severe degradation of the seismic image. To meet the challenges of speeding up imaging workflows and retaining accurate salt picks, we evaluate three deep learning approaches: a 2D window-based convolutional neural network, a 3D window-based convolutional neural network, and finally a 2D “U-Net” approach. A 3D seismic volume from the deep-water field Julia in the Gulf of Mexico was used to test these approaches. The Julia field has complex salt structures with overhangs and inclusions, and the thickness of salt can reach up to 5 km. The U-Net architecture proved to be the most accurate of the three methods tested, predicting the placement of salt at 98% accuracy, as compared to the human interpretation. Beyond accuracy, U-Net also proved to be the fastest, requiring only 3.5 hours to predict salt on the 3D seismic volume. The results presented here along with other recent studies of deep learning for salt interpretation represent a clear shift in the seismic interpretation workflow. Show less

I present an application of a high-resolution subsurface imaging technique known as “full-waveform inversion” (FWI) to a vibroseis seismic dataset from eastern Ohio, USA. The data were collected over a crooked line with rough topography, 3.5 km maximum offsets, and no significant frequency content below 12 Hz. These parameters present challenges to obtaining quality images from FWI. The use of a preconditioner–the ‘scaled-Sobolev preconditioner’ (SSP - Zuberi and Pratt, 2017)–on the gradient of… Show more

I present an application of a high-resolution subsurface imaging technique known as “full-waveform inversion” (FWI) to a vibroseis seismic dataset from eastern Ohio, USA. The data were collected over a crooked line with rough topography, 3.5 km maximum offsets, and no significant frequency content below 12 Hz. These parameters present challenges to obtaining quality images from FWI. The use of a preconditioner–the ‘scaled-Sobolev preconditioner’ (SSP - Zuberi and Pratt, 2017)–on the gradient of the misfit functional was key to obtaining low wavenumbers without discarding high wavenumbers. The results represent the first successful application of FWI with the SSP to a field dataset, with a high-resolution image that generally matches the trends of the Big Injun sand and Berea sandstone layers at the survey location. The novel FWI results confirm the absence of small scale structure (including the lack of visible faults) in the first 0.66 km. Show less

We present an application of seismic full-waveform inversion (FWI) with scaled-Sobolev preconditioning (SSP - after Zuberi and Pratt, 2016) to a field seismic dataset. The data were collected over a crooked line with rough topography, using only short offsets, and no low frequency content. As is often the case with land data, this presents a number of challenges for processing with FWI. The reservoir of interest is the Utica Shale in eastern Ohio, USA, which sits at depths between 2.2 km and… Show more

We present an application of seismic full-waveform inversion (FWI) with scaled-Sobolev preconditioning (SSP - after Zuberi and Pratt, 2016) to a field seismic dataset. The data were collected over a crooked line with rough topography, using only short offsets, and no low frequency content. As is often the case with land data, this presents a number of challenges for processing with FWI. The reservoir of interest is the Utica Shale in eastern Ohio, USA, which sits at depths between 2.2 km and 2.5 km in the study area. We limited our inversions to approximately 0.6 km in depth, due to the narrow range of offsets. Constraining the velocity structure in the very near surface is essential to recovering the velocity structure at greater depths through subsequent migration processing. The FWI results are validated by a comparison of forward-modelled data to field data, and by a scrutiny of the coherencies of the recovered source signatures. Show less