Ehsan Zabihi Naeini
Ehsan Zabihi Naeini has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Sagar Singh
Colorado School of Mines 1 , Center for Wave Phenomena, Golden, Colorado 80401, USA.;.
4 shared publications - Ilya Tsvankin
Center for Wave Phenomena, Department of Geophysics, Colorado School of Mines, Golden, CO 80401, USA
3 shared publications - Ali Gholami
Institute of Geophysics Polish Academy of Sciences Warsaw Poland
1 shared publication - Alvaro L. G. A. Coutinho
Department of Civil Engineering, COPPE/Federal University of Rio de Janeiro, Rio de Janeiro, Brazil
1 shared publication - Behzad Alaei
Earth Science Analytics 2 , Lysaker, Norway and , Department of Geosciences, Oslo, Norway and , Department of Geosciences, Oslo, Norway
1 shared publication - Dave Hale
Colorado School of Mines 1 , Golden, Colorado, .;.
1 shared publication
Latest Publications
- 22 April 2022
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 87 (4) , IM157-IM176
A comprehensive study of calibration and uncertainty quantification for Bayesian convolutional neural networks — An application to seismic data
Deep neural networks offer numerous potential applications across geoscience; for example, one could argue that they are the state-of-the-art method for predicting faults in seismic data sets. In quantitative reservoir characterization workflows, it is common to incorporate the uncertainty of predictions. Thus, such subsurface models should provide calibrated probabilities and the associated uncertainties in their predictions. It has been shown that popular deep learning-based models often are miscalibrated, and due to their deterministic nature, they provide no means to interpret the uncertainty of their predictions. We compare three different approaches for obtaining probabilistic models based on convolutional neural networks in a Bayesian formalism, namely, deep ensembles, concrete dropout, and stochastic weight averaging Gaussian (SWAG). These methods are consistently applied to fault detection case studies in which deep ensembles use independently trained models to provide fault probabilities and concrete dropout represents an extension to the popular dropout technique to approximate Bayesian neural networks. Finally, we apply SWAG, a recent method that is based on the Bayesian inference equivalence of minibatch stochastic gradient descent. We provide quantitative results in terms of model calibration and uncertainty representation, as well as qualitative results on the synthetic and real seismic data sets. Our results indicate that the approximate Bayesian methods, concrete dropout and SWAG, provide well-calibrated predictions and uncertainty attributes at a lower computational cost compared to the baseline deep ensemble approach. The resulting uncertainties also offer a possibility to further improve the model performance, as well as enhancing the interpretability of the models.
- 22 April 2022
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 87 (4) , IM157-IM176
A comprehensive study of calibration and uncertainty quantification for Bayesian convolutional neural networks — An application to seismic data
Deep neural networks offer numerous potential applications across geoscience; for example, one could argue that they are the state-of-the-art method for predicting faults in seismic data sets. In quantitative reservoir characterization workflows, it is common to incorporate the uncertainty of predictions. Thus, such subsurface models should provide calibrated probabilities and the associated uncertainties in their predictions. It has been shown that popular deep learning-based models often are miscalibrated, and due to their deterministic nature, they provide no means to interpret the uncertainty of their predictions. We compare three different approaches for obtaining probabilistic models based on convolutional neural networks in a Bayesian formalism, namely, deep ensembles, concrete dropout, and stochastic weight averaging Gaussian (SWAG). These methods are consistently applied to fault detection case studies in which deep ensembles use independently trained models to provide fault probabilities and concrete dropout represents an extension to the popular dropout technique to approximate Bayesian neural networks. Finally, we apply SWAG, a recent method that is based on the Bayesian inference equivalence of minibatch stochastic gradient descent. We provide quantitative results in terms of model calibration and uncertainty representation, as well as qualitative results on the synthetic and real seismic data sets. Our results indicate that the approximate Bayesian methods, concrete dropout and SWAG, provide well-calibrated predictions and uncertainty attributes at a lower computational cost compared to the baseline deep ensemble approach. The resulting uncertainties also offer a possibility to further improve the model performance, as well as enhancing the interpretability of the models.
- 24 January 2022
- Published by Society of Exploration Geophysicists in Interpretation
- Vol. 10 (2) , T279-T290
Facies prediction with Bayesian inference: Application of supervised and semisupervised deep learning
Accurate delineation of geologic facies and determination of live fluids from seismic reflection data is crucial for reservoir characterization during petroleum exploration. Facies classification or fluid identification is often done manually by an experienced interpreter, which makes this process subjective, laborious, and time-consuming. Several machine-learning models have been proposed to automate multiclass facies segmentation, but significant practical challenges (e.g., limited scope of labels for training purposes, skewed data distribution, inefficient performance evaluation metrics, etc) still remain. We present supervised and semisupervised Bayesian deep-learning methodologies to improve analysis of seismic facies depending on the scope of the labeled data. The developed networks reliably predict facies distribution using seismic reflection data and estimate the corresponding uncertainty. Therefore, they provide more consistent and meaningful information for seismic interpretation than commonly used deterministic approaches. We apply the proposed deep-learning models to field data from the North Sea to demonstrate the generalized-prediction capabilities of our methodology. In the case of sufficient availability of manually interpreted labels (or facies), the supervised learning model accurately recovers the facies distribution. When the amount of the interpreted labels is limited, we efficiently apply the semisupervised algorithm to avoid overfitting.
- 24 January 2022
- Published by Society of Exploration Geophysicists in Interpretation
- Vol. 10 (2) , T279-T290
Facies prediction with Bayesian inference: Application of supervised and semisupervised deep learning
Accurate delineation of geologic facies and determination of live fluids from seismic reflection data is crucial for reservoir characterization during petroleum exploration. Facies classification or fluid identification is often done manually by an experienced interpreter, which makes this process subjective, laborious, and time-consuming. Several machine-learning models have been proposed to automate multiclass facies segmentation, but significant practical challenges (e.g., limited scope of labels for training purposes, skewed data distribution, inefficient performance evaluation metrics, etc) still remain. We present supervised and semisupervised Bayesian deep-learning methodologies to improve analysis of seismic facies depending on the scope of the labeled data. The developed networks reliably predict facies distribution using seismic reflection data and estimate the corresponding uncertainty. Therefore, they provide more consistent and meaningful information for seismic interpretation than commonly used deterministic approaches. We apply the proposed deep-learning models to field data from the North Sea to demonstrate the generalized-prediction capabilities of our methodology. In the case of sufficient availability of manually interpreted labels (or facies), the supervised learning model accurately recovers the facies distribution. When the amount of the interpreted labels is limited, we efficiently apply the semisupervised algorithm to avoid overfitting.
- 11 May 2021
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 86 (4) , R589-R602
Elastic FWI for orthorhombic media with lithologic constraints applied via machine learning
Full-waveform inversion (FWI) of 3D wide-azimuth data for elastic orthorhombic media suffers from parameter trade-offs which cannot be overcome without constraining the model-updating procedure. We present an FWI methodology that incorporates geologic constraints to reduce the inversion nonlinearity and increase the resolution of parameter estimation for orthorhombic models. These constraints are obtained from well logs, which can provide rock-physics relationships for different geologic facies. Because the locations of the available well logs are usually sparse, a supervised machine-learning (ML) algorithm (Support Vector Machine) is employed to account for lateral heterogeneity in building the lithologic constraints. The advantages of the facies-based FWI are demonstrated on the modified SEG-EAGE 3D overthrust model, which is made orthorhombic with the symmetry planes that coincide with the Cartesian coordinate planes. We employ a velocity-based parameterization, whose suitability for FWI was studied using the radiation-pattern analysis. Application of the facies-based constraints substantially increases the resolution of the P- and S-wave vertical velocities (VP0, VS0, and VS1) and, therefore, of the depth scale of the model. Improvements are also observed for the P-wave horizontal and normal-moveout velocities (VP1, VP2, Vnmo,1, and Vnmo,2) and the S-wave horizontal velocity VS2. However, the velocity Vnmo,3 that depends on Tsvankin’s parameter δ(3) defined in the horizontal plane is not well recovered from the surface data. On the whole, the developed algorithm achieves a much higher spatial resolution compared to unconstrained FWI, even in the absence of recorded frequencies below 2 Hz.
- 11 May 2021
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 86 (4) , R589-R602
Elastic FWI for orthorhombic media with lithologic constraints applied via machine learning
Full-waveform inversion (FWI) of 3D wide-azimuth data for elastic orthorhombic media suffers from parameter trade-offs which cannot be overcome without constraining the model-updating procedure. We present an FWI methodology that incorporates geologic constraints to reduce the inversion nonlinearity and increase the resolution of parameter estimation for orthorhombic models. These constraints are obtained from well logs, which can provide rock-physics relationships for different geologic facies. Because the locations of the available well logs are usually sparse, a supervised machine-learning (ML) algorithm (Support Vector Machine) is employed to account for lateral heterogeneity in building the lithologic constraints. The advantages of the facies-based FWI are demonstrated on the modified SEG-EAGE 3D overthrust model, which is made orthorhombic with the symmetry planes that coincide with the Cartesian coordinate planes. We employ a velocity-based parameterization, whose suitability for FWI was studied using the radiation-pattern analysis. Application of the facies-based constraints substantially increases the resolution of the P- and S-wave vertical velocities (VP0, VS0, and VS1) and, therefore, of the depth scale of the model. Improvements are also observed for the P-wave horizontal and normal-moveout velocities (VP1, VP2, Vnmo,1, and Vnmo,2) and the S-wave horizontal velocity VS2. However, the velocity Vnmo,3 that depends on Tsvankin’s parameter δ(3) defined in the horizontal plane is not well recovered from the surface data. On the whole, the developed algorithm achieves a much higher spatial resolution compared to unconstrained FWI, even in the absence of recorded frequencies below 2 Hz.
- 1 February 2021
- Published by Society of Exploration Geophysicists in Interpretation
- Vol. 9 (1) , SA17-SA22
Cloud computing in geoscience: Mysteries, miseries, and benefits
Cloud computing has become an integral part of our daily life and work. In the oil and gas industry, cloud computing is becoming increasingly attractive to experts, operators and software companies...
- 1 February 2021
- Published by Society of Exploration Geophysicists in Interpretation
- Vol. 9 (1) , SA17-SA22
Cloud computing in geoscience: Mysteries, miseries, and benefits
Cloud computing has become an integral part of our daily life and work. In the oil and gas industry, cloud computing is becoming increasingly attractive to experts, operators and software companies...
- 1 November 2020
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 85 (6) , R553-R563
Full-waveform inversion with borehole constraints for elastic VTI media
The nonlinearity of full-waveform inversion (FWI) and parameter trade-offs can prevent convergence toward the actual model, especially for elastic anisotropic media. The problems with parameter updating become particularly severe if ultra-low-frequency seismic data are unavailable, and the initial model is not sufficiently accurate. We introduce a robust way to constrain the inversion workflow using borehole information obtained from well logs. These constraints are included in the form of rock-physics relationships for different geologic facies (e.g., shale, sand, salt, and limestone). We develop a multiscale FWI algorithm for transversely isotropic media with a vertical symmetry axis (VTI media) that incorporates facies information through a regularization term in the objective function. That term is updated during the inversion by using the models obtained at the previous inversion stage. To account for lateral heterogeneity between sparse borehole locations, we use an image-guided smoothing algorithm. Numerical testing for structurally complex anisotropic media demonstrates that the facies-based constraints may ensure the convergence of the objective function towards the global minimum in the absence of ultra-low-frequency data and for simple (even 1D) initial models. We test the algorithm on clean data and on surface records contaminated by Gaussian noise. The algorithm also produces a high-resolution facies model, which should be instrumental in reservoir characterization.
- 1 November 2020
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 85 (6) , R553-R563
Full-waveform inversion with borehole constraints for elastic VTI media
The nonlinearity of full-waveform inversion (FWI) and parameter trade-offs can prevent convergence toward the actual model, especially for elastic anisotropic media. The problems with parameter updating become particularly severe if ultra-low-frequency seismic data are unavailable, and the initial model is not sufficiently accurate. We introduce a robust way to constrain the inversion workflow using borehole information obtained from well logs. These constraints are included in the form of rock-physics relationships for different geologic facies (e.g., shale, sand, salt, and limestone). We develop a multiscale FWI algorithm for transversely isotropic media with a vertical symmetry axis (VTI media) that incorporates facies information through a regularization term in the objective function. That term is updated during the inversion by using the models obtained at the previous inversion stage. To account for lateral heterogeneity between sparse borehole locations, we use an image-guided smoothing algorithm. Numerical testing for structurally complex anisotropic media demonstrates that the facies-based constraints may ensure the convergence of the objective function towards the global minimum in the absence of ultra-low-frequency data and for simple (even 1D) initial models. We test the algorithm on clean data and on surface records contaminated by Gaussian noise. The algorithm also produces a high-resolution facies model, which should be instrumental in reservoir characterization.
- 17 March 2020
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 85 (4) , WA159-WA171
Missing well log prediction using convolutional long short-term memory network
Reservoir characterization involves integration of different types of data to understand the subsurface rock properties. To incorporate multiple well log types into reservoir studies, estimating missing logs is an essential step. We have developed a method to estimate missing well logs by using a bidirectional convolutional long short-term memory (bidirectional ConvLSTM) cascaded with fully connected neural networks. We train the model on 177 wells from mature areas of the UK continental shelf (UKCS). We test the trained model on one blind well from UKCS, three wells from the Volve field in the Norwegian continental shelf, one well from the Penobscot field in the Scotian shelf offshore Canada, and one well from the Teapot Dome data set in Wyoming. The method takes into account the depth trend and the local shape of logs by using ConvLSTM architecture. The method is examined on sonic log prediction and can produce an accurate prediction of sonic logs from gamma-ray, density, and neutron porosity logs. The advantages of our method are that it is not applied on an interval by interval basis like rock-physics-based methods and it also outputs the uncertainties facilitated by dropout layers and Monte Carlo sampling at inference time.
- 17 March 2020
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 85 (4) , WA159-WA171
Missing well log prediction using convolutional long short-term memory network
Reservoir characterization involves integration of different types of data to understand the subsurface rock properties. To incorporate multiple well log types into reservoir studies, estimating missing logs is an essential step. We have developed a method to estimate missing well logs by using a bidirectional convolutional long short-term memory (bidirectional ConvLSTM) cascaded with fully connected neural networks. We train the model on 177 wells from mature areas of the UK continental shelf (UKCS). We test the trained model on one blind well from UKCS, three wells from the Volve field in the Norwegian continental shelf, one well from the Penobscot field in the Scotian shelf offshore Canada, and one well from the Teapot Dome data set in Wyoming. The method takes into account the depth trend and the local shape of logs by using ConvLSTM architecture. The method is examined on sonic log prediction and can produce an accurate prediction of sonic logs from gamma-ray, density, and neutron porosity logs. The advantages of our method are that it is not applied on an interval by interval basis like rock-physics-based methods and it also outputs the uncertainties facilitated by dropout layers and Monte Carlo sampling at inference time.
- 1 October 2019
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 38 (10) , 770-779
Facies-based seismic inversion for a field development plan
Field appraisal and development plans aim to provide the best technical solution for optimizing hydrocarbon production and require integration between various disciplines including geology, geophysics, engineering, well planning, and environmental sciences. Seismic inversion could provide one essential component for reservoir modeling in support of appraisal and development evaluations. Therefore, it is important to quantitatively assess all of the possibilities and uncertainties involved in reservoir definition and extension. A probabilistic facies-based seismic inversion method has been utilized to achieve this goal in a recent Central North Sea discovery. The probabilistic nature of the inversion allows computation of various scenarios. We categorically selected, among others, most likely, optimistic, and pessimistic scenarios based on prior knowledge and calibration at the wells. Then, we performed a statistical analysis of all of the scenarios to identify the uncertainties. We also performed a postinversion forward-modeling study to assess uncertainties that may be related to thin layers of subseismic resolution.
- 1 October 2019
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 38 (10) , 770-779
Facies-based seismic inversion for a field development plan
Field appraisal and development plans aim to provide the best technical solution for optimizing hydrocarbon production and require integration between various disciplines including geology, geophysics, engineering, well planning, and environmental sciences. Seismic inversion could provide one essential component for reservoir modeling in support of appraisal and development evaluations. Therefore, it is important to quantitatively assess all of the possibilities and uncertainties involved in reservoir definition and extension. A probabilistic facies-based seismic inversion method has been utilized to achieve this goal in a recent Central North Sea discovery. The probabilistic nature of the inversion allows computation of various scenarios. We categorically selected, among others, most likely, optimistic, and pessimistic scenarios based on prior knowledge and calibration at the wells. Then, we performed a statistical analysis of all of the scenarios to identify the uncertainties. We also performed a postinversion forward-modeling study to assess uncertainties that may be related to thin layers of subseismic resolution.
- 1 May 2019
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 84 (3) , R311-R320
3D Dix inversion using bound-constrained total variation regularization
Given the ill-conditioned nature of Dix inversion, the resultant Dix interval-velocity field is often unrealistic, noisy, and highly dependent on the quality of the provided root-mean-square velocities. While the classic least-squares regularization techniques, e.g., various forms of Tikhonov regularization, lead to somewhat suboptimal stability, we formulated the Dix inversion as a new constrained optimization problem. This enables one to incorporate prior knowledge as soft and/or hard bounds for the optimization, effectively treating it as a denoising problem. The solution to the problem is achieved by a bound-constrained total variation (TV) regularization. TV regularization has the advantage of being able to recover the discontinuities in the model, but it often comes with a large memory and compute requirements. Therefore, we have developed a simple and memory-efficient algorithm using iterative refinement strategy. The quality of the new algorithm is also cross-examined against different strategies, which are currently used in practice. Overall, we observe that the proposed method outperforms classic Dix inversion methods on the synthetic and real data examples.
- 1 May 2019
- Published by Society of Exploration Geophysicists in Geophysics
- Vol. 84 (3) , R311-R320
3D Dix inversion using bound-constrained total variation regularization
Given the ill-conditioned nature of Dix inversion, the resultant Dix interval-velocity field is often unrealistic, noisy, and highly dependent on the quality of the provided root-mean-square velocities. While the classic least-squares regularization techniques, e.g., various forms of Tikhonov regularization, lead to somewhat suboptimal stability, we formulated the Dix inversion as a new constrained optimization problem. This enables one to incorporate prior knowledge as soft and/or hard bounds for the optimization, effectively treating it as a denoising problem. The solution to the problem is achieved by a bound-constrained total variation (TV) regularization. TV regularization has the advantage of being able to recover the discontinuities in the model, but it often comes with a large memory and compute requirements. Therefore, we have developed a simple and memory-efficient algorithm using iterative refinement strategy. The quality of the new algorithm is also cross-examined against different strategies, which are currently used in practice. Overall, we observe that the proposed method outperforms classic Dix inversion methods on the synthetic and real data examples.
- 1 January 2019
- Published by EAGE Publications bv
- p. 1-5
3D Pore Pressure and Geomechanics: Work Smarter and Faster Integrating Geoscience with Machine Learning
- 1 January 2019
- Published by EAGE Publications bv
- p. 1-5
3D Pore Pressure and Geomechanics: Work Smarter and Faster Integrating Geoscience with Machine Learning
- 1 January 2019
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 38 (1) , 53-59
An integrated deep learning solution for petrophysics, pore pressure, and geomechanics property prediction
In unconventional plays, wells are drilled at an unprecedented rate. This, together with technical challenges in terms of complex stratigraphy, multiple play types, variable rock properties, and various elements of pore pressure, geomechanics, fracturing, and diagenesis, calls for more sophisticated, faster, consistent, and wider ranging analytical tools. Given the scale of the work — i.e., the number of wells — performing classical workflows for petrophysics, pore pressure, and geomechanics prediction can be impractical (if not impossible) due to turnaround considerations. Also such workflows might not use any preexisting regional studies efficiently. In principle, a machine learning approach can mitigate these shortcomings. We show that a supervised deep neural network approach can be an alternative innovative tool for petrophysical, pore pressure, and geomechanics analysis enabling the use of all the previously interpreted data to devise solutions that simultaneously integrate wide-ranging wellbore and wireline logs. Beyond that, a similar approach is taken to predict a certain number of attributes solely from seismically derived properties, which allows one to compute volumetric models. The application of such an algorithm is shown on a Permian case study in which the automatic neural-network-based algorithms achieve reasonable accuracy in a fraction of the time.
- 1 January 2019
- Published by Society of Exploration Geophysicists in The Leading Edge
- Vol. 38 (1) , 53-59
An integrated deep learning solution for petrophysics, pore pressure, and geomechanics property prediction
In unconventional plays, wells are drilled at an unprecedented rate. This, together with technical challenges in terms of complex stratigraphy, multiple play types, variable rock properties, and various elements of pore pressure, geomechanics, fracturing, and diagenesis, calls for more sophisticated, faster, consistent, and wider ranging analytical tools. Given the scale of the work — i.e., the number of wells — performing classical workflows for petrophysics, pore pressure, and geomechanics prediction can be impractical (if not impossible) due to turnaround considerations. Also such workflows might not use any preexisting regional studies efficiently. In principle, a machine learning approach can mitigate these shortcomings. We show that a supervised deep neural network approach can be an alternative innovative tool for petrophysical, pore pressure, and geomechanics analysis enabling the use of all the previously interpreted data to devise solutions that simultaneously integrate wide-ranging wellbore and wireline logs. Beyond that, a similar approach is taken to predict a certain number of attributes solely from seismically derived properties, which allows one to compute volumetric models. The application of such an algorithm is shown on a Permian case study in which the automatic neural-network-based algorithms achieve reasonable accuracy in a fraction of the time.