Earth Science and Engineering Program
For more information visit: https://pse.kaust.edu.sa/study/academicprograms/earthscienceandengineering/Pages/home.aspx
Recent Submissions

Bulk and Interfacial Properties of the Decane + Water System in the Presence of Methane, Carbon Dioxide, and Their Mixture(The Journal of Physical Chemistry B, American Chemical Society (ACS), 20201016) [Article]Molecular dynamics simulations are carried out to study the twophase behavior of the ndecane + water system in the presence of methane, carbon dioxide, and their mixture under reservoir conditions. The simulation studies were complemented by theoretical modeling using the perturbedchain statistical associating fluid theory (PCSAFT) equation of state (EoS) and density gradient theory. Our results show that the presence of methane and carbon dioxide decreases the interfacial tension (IFT) of the decane + water system. In general, the IFT increases with increasing pressure and decreasing temperature for the methane + decane + water and carbon dioxide + decane + water systems, similar to what has been found for the corresponding decane + water system. The most important finding of this study is that the presence of carbon dioxide decreases the IFT of the methane + decane + water system. The atomic density profiles provide evidence of the local accumulation of methane and carbon dioxide at the interface, in most of the studied systems. The results of this study show the preferential dissolution in the waterrich phase and enrichment at the interface for carbon dioxide in the methane + carbon dioxide + decane + water system. This indicates the preferential interaction of water with carbon dioxide relative to methane and decane. Notably, there is an enrichment of the interface by decane at high mole fractions of methane in the methane/decanerich or methane/carbon dioxide/decanerich phase. Overall, the solubility of methane and carbon dioxide in the waterrich phase increases with increasing pressure and temperature. Additionally, we find that the overall performance of the PCSAFT EoS and the cubicplusassociation EoS is similar with respect to the calculation of bulk and interfacial properties of these systems.

Organic carbon export and loss rates in the Red Sea(Global Biogeochemical Cycles, American Geophysical Union (AGU), 20201014) [Article]The export and fate of organic carbon in the mesopelagic zone are still poorly understood and quantified due to lack of observations. We exploited data from a BGCArgo float that was deployed in the Red Sea to study how a warm and hypoxic environment can affect the fate of the organic carbon in the ocean’s interior. We observed that only 10% of the particulate organic carbon (POC) exported survived at depth due to remineralization processes in the upper mesopelagic zone. We also found that POC exported was rapidly degraded in a first stage and slowly in a second one, which may be dependent on the palatability of the organic matter. We observed that AOUbased loss rates (a proxy of the remineralization of total organic matter) were significantly higher than the POCbased loss rates, likely because changes in AOU are mainly attributed to changes in dissolved organic carbon. Finally, we showed that POC and AOUbased loss rates could be expressed as a function of temperature and oxygen concentration. These findings advance our understanding of the biological carbon pump and mesopelagic ecosystem.

Stability theory of nanofluid over an exponentially stretching cylindrical surface containing microorganisms.(Scientific reports, Springer Science and Business Media LLC, 20201013) [Article]This research is emphasized to describe the stability analysis in the form of dual solution of the flow and heat analysis on nanofluid over an exponential stretching cylindrical surface containing microorganisms. The research is also implemented to manifest the dual profiles of velocity, temperature and nanoparticle concentration in the effect of velocity ratio parameter ([Formula: see text]). Living microorganisms' cell are mixed into the nanofluid to neglect the unstable condition of nano type particles. The governing equations are transformed to nonlinear ordinary differential equations with respect to pertinent boundary conditions by using similarity transformation. The significant differential equations are solved using build in function bvp4c in MATLAB. It is seen that the solution is not unique for vertical stretching sheet. This research is reached to excellent argument when found results are compared with available result. It is noticed that dual results are obtained demanding on critical value ([Formula: see text]), the meanings are indicated at these critical values both solutions are connected and behind these critical value boundary layer separates thus the solution are not stable.

Seismic inversion by multidimensional Newtonian machine learning(Society of Exploration Geophysicists, 20201001) [Conference Paper]Newtonian machine learning (NML) inversion has been shown to accurately recover the lowtointermediate wavenumber information of subsurface velocity models. This method uses the waveequation inversion kernel to invert the skeletonized data that is automatically learned by an autoencoder. The skeletonised data is a onedimensional latentspace representation of the seismic trace. However, for a complicated dataset, the decoded waveform could lose some details if the latent space dimension is set to one, which leads to a lowresolution NML tomogram. To mitigate this problem, an autoencoder with a higher dimensional latent space is needed to encode and decode the seismic data. In this paper, we present a wave equation inversion that inverts the multidimensional latent variables of an autoencoder for the subsurface velocity model. The multivariable implicit function theorem is used to determine the perturbation of the multidimensional skeletonised data with respect to the velocity perturbations. In this case, each dimension of the latent variable is characterized one gradient and the velocity model is updated by the weighted sum of all these gradients. Numerical results suggest that the multidimensional NML inverted result can achieve a higher resolution in the tomogram compared to the conventional single dimensional NML inversion.

Fullwaveform inversion with an exponential filter in wavenumber domain(Society of Exploration Geophysicists, 20201001) [Conference Paper]The gradient of fullwaveform inversion (FWI) can be decomposed into a tomographic part, which is smooth (lowwavenumbers) and an image part, which is more sharp (highwavenumbers). Most of the FWI procedures need to update the smooth part first and gradually include the details to mitigate the cycleskipping problem. The scattering angle filter for the gradient is one way to isolate the lowwavenumber part of the gradient at the early stage. However, it could be costly because of the extensions necessary for precise scattering angle control. Due to the relationship between the wavenumber and the scattering angle, the large scattering angles correspond to the low wavenumber component, which is usually what we want to update in our initial inversion steps. Thus, we use an exponential filter in the wavenumber domain and update the model in the wavenumber domain from low to high. The numerical result indicates that the inversion benefits from the wavenumber domain filter.

Extrapolating lowfrequency prestack land data with deep learning(Society of Exploration Geophysicists, 20201001) [Conference Paper]Missing lowfrequency content in seismic data is a common challenge for seismic inversion. Long wavelengths are necessary to reveal large structures in the subsurface and to build an acceptable starting point for later iterations of fullwaveform inversion (FWI). Highfrequency land seismic data are particularly challenging due to the elastic nature of the Earth contrasting with acoustic air at the typically rugged free surface, which makes the use of low frequencies even more vital to the inversion. We propose a supervised deep learning framework for bandwidth extrapolation of prestack elastic data in the time domain. We utilize a Convolutional Neural Network (CNN) with a UNetinspired architecture to convert portions of bandlimited shot gathers from 515 Hz to 05 Hz band. In the synthetic experiment, we train the network on 192x192 patches of wavefields simulated for different crosssections of the elastic SEAM Arid model with freesurface. Then, we test the network on unseen shot gathers from the same model to demonstrate the viability of the approach. The results show promise for future field data applications.

Predict passive seismic events with a convolutional neural network(Society of Exploration Geophysicists, 20201001) [Conference Paper]The ample size of time lapse data requires tremendous event detection and source locating capabilities, especially in areas like shale gas exploration regions where a large number of passive seismic events are often recorded. In many cases, the realtime monitoring and locating of these events are essential to production decisions. Conventional methods face considerable drawbacks. For example, traveltime based methods require traveltime picking; migration methods, on the other hand, require many wavefield modeling applications. These human interaction based pickings or wavefield simulations face severe issues when too many passive sources need to be located, which is common in shale gas explorations. A purely automatic method with no human interactions and less computational cost is necessary. Recently, machine learning has been utilized for this task, whether to identity passive seismic events, or to locate their sources once they are identified and picked. We propose to use an artificial neural network to directly map seismic data, without any picking, to locations of potential seismic events. A visual geometry group (VGG) neural network is trained on synthetic acoustic data, corresponding to known sources, to provide predictions of new passive source locations, and other source features such as source wavelet peak frequencies, amplitudes and the number of sources within the data segment. To reduce the size of the inputtothenetwork seismic data, we correlate the traces with the central trace to allow the network to focus on the curvature of the input data. We train the network to handle both single and double events that might be included in the correlation window. An initial application of the approach on a simple V(z) model demonstrates its potential.

Targetoriented timelapse waveform inversion using a deep learningassisted regularization(Society of Exploration Geophysicists, 20201001) [Conference Paper]Detection of the property changes in the subsurface during production is important, yet it is also very challenging considering that these changes are often very settle. The quantitative evaluation of the subsurface property obtained by full waveform inversion (FWI) allows for better monitoring of these timelapse changes. However, highresolution inversion is usually accompanied with a large computational burden. Besides, the resolution of inversion is limited by the bandwidth and aperture of timelapse seismic data. We apply a targetoriented strategy through seismic redatuming to reduce the computational cost by focusing our highresolution delineation on a relatively small target zone. The redatuming technique enables retrieving timelapse virtual data for the targetoriented inversion. Considering the injection and production wells are often present in the target zone, we can incorporate the well information to the timelapse inversion by using regularization to complement the resolution and illumination at the reservoir. We use a deep neural network (DNN) to learn the mathematical relationship between the inverted model and the facies interpreted from well logs. The trained network is employed to map the property changes extracted from the wells to the target inversion domain. We then perform another timelapse inversion, in which we fit the predicted data difference to the redatumed one from observation, as well as fit the model to the predicted velocity changes. The numerical results demonstrate that the proposed method is capable of inverting for the timelapse changes effectively in the target zone by incorporating the learned model information from well logs.

Targetoriented timelapse waveform inversion using redatumed data: Feasibility and robustness(Society of Exploration Geophysicists, 20201001) [Conference Paper]Seismic monitoring of the changes in the subsurface induced by various types of injections into reservoirs is important, yet challenging. Timelapse waveform inversion can retrieve quantitative estimates of subsurface property changes. Considering that property changes usually occur in a limited region rather than the whole subsurface, we present a targetoriented timelapse waveform inversion method, which allows for dynamic monitoring of the target of interest. We employ a redatuming technique to produce virtual data at a desired datum level for the targetoriented inversion. Given the redatumed timelapse data, the property changes can be quantitatively estimated from the data difference for the virtual survey using a doubledifference waveform inversion (DDWI). In the numerical examples, the dependence of the inversion performance on the quality of overburden model and its robustness to nonrepeatable acquisition survey and random noise is investigated. The numerical results demonstrate that the inversion method is capable of recovering the timelapse changes reasonably well under some challenging circumstances. We will show field data examples at the conference.

Machine learned Green's functions that approximately satisfy the wave equation(Society of Exploration Geophysicists, 20201001) [Conference Paper]Green’s functions are wavefield solutions for a particular point source. They form basis functions to build wavefields for modeling and inversion. However, calculating Green’s functions are both costly and memory intensive. We formulate frequencydomain machinelearned Green’s functions that are represented by neural networks (NN). This NN outputs a complex number (two values representing the real and imaginary part) for the scattered Green’s function at a location in space for a specific source location (both locations are input to the network). Considering a background homogeneous medium admitting an analytical Green’s function solution, the network is trained by fitting the output perturbed Green’s function and its derivatives to the wave equation expressed in terms of the perturbed Green’s function. The derivatives are calculated through the concept of automatic differentiation. In this case, the background Green’s function absorbs the point source singularity, which will allow us to train the network using random points over space and source location using a uniform distribution. Thus, feeding a reasonable number of random points from the model space will ultimately train a fully connected 8layer deep neural network, to predict the scattered Green’s function. Initial tests on part of the simple layered model (extracted from the left side of the Marmousi model) with sources on the surface demonstrate the successful training of the NN for this application. Using the trained NN model for the Marmousi as an initial NN model for solving for the scattered Green’s function for a 2D slice from the Sigsbee model helped the NN converge faster to a reasonable solution

Waveequation migration velocity analysis via the optimaltransportbased objective function(Society of Exploration Geophysicists, 20201001) [Conference Paper]Wave equation migration velocity analysis (WEMVA) builds a kinematically accurate macro velocity model containing the largescale structures of the model for seismic imaging or full waveform inversion (FWI). Differential semblance optimization (DSO) formulates the misfit function in the subsurface domain by applying a penalty to the unfocused subsurfaceoffset gathers or the unflatness of the angle gathers. Such penalty applied by DSO leads to gradients with strong artifacts and thus converges slower. Here, we propose to formulate the misfit function using the optimal transport (OT) between neighbouring traces in angle domain common image gathers (ADCIGs). Specially, we measure the unflatness of the gathers by comparing the Wasserstein distance of adjacent traces. The proposed objective function is expected to be minimum for the correct velocity when the angle gathers are flat. Numerical examples of a twolayer and Marmousi models demonstrate the validity of the new method for estimating the macro velocity model.

A new Swave free acoustic approximation for the transversely isotropic media with a vertical symmetry axis(Society of Exploration Geophysicists, 20201001) [Conference Paper]To derive a totally Swave free acoustic approximation, we propose a new acoustic approximation for pure Pwaves that is totally free of Swave artifacts in the homogenous VTI model. To keep the Swave velocity equal to zero, we formulate the vertical Swave velocity to be a function of the model parameters, rather than simply setting it to zero. Then, the corresponding Pwave phase and group velocities for the new acoustic approximation are derived. For this new acoustic approximation, the kinematics is described by a new eikonal equation for pure Pwave propagation, which defines the new vertical slowness for the Pwaves. The accuracy of our new Pwave acoustic approximation is tested on numerical examples. We find that the accuracy of our new acoustic approximation is as good as the original one for the traveltime and the relative geometrical spreading. Therefore, the Swavefree acoustic approximation could be further applied in seismic processing that requires pure Pwave data.

Elastic nearsurface model estimation from full waveforms by deep learning(Society of Exploration Geophysicists, 20201001) [Conference Paper]Strong nearsurface heterogeneity poses a major challenge for seismic imaging of deep targets in arid environments. Inspired by the initial success of deep learning applications to inverse problems, we investigate the possibility of building nearsurface models directly from raw elastic data including surface and body waves in arid conditions. Namely, we train a convolutional neural network to map the data into the model directly in a supervised way on a part SEAM Arid synthetic dataset and evaluate its performance on a different part of the same dataset. The main feature of our approach is that we estimate the model as a set of 1D vertical velocity profiles, utilizing relevant subsets of input data from neighboring locations. This effectively reduces the data and label spaces for a more practical neural network application.

Velocity model building by deep learning: From general synthetics to field data application(Society of Exploration Geophysicists, 20201001) [Conference Paper]Velocity model building is not straightforward in geologically complex environments. We train a convolutional neural network (CNN) to map full wavefields to smooth subsurface parameter distributions to address the problem. Specifically, cubes of neighboring CMP gathers are mapped into in 1D vertical profiles to simplify the training phase and to make it easier to utilize well logs in future applications. We train the CNN using a total of one hundred thousand random subsurface models generated onthefly and the corresponding synthetic data. The application of the trained CNN on synthetic and real data admitted reasonably accurate models representing mostly the low wavenumber features of the true models.

Joint seismic and electromagnetic inversion for reservoir mapping using a deep learning aided featureoriented approach(Society of Exploration Geophysicists, 20201001) [Conference Paper]As more and more types of geophysical measurements informing about different characteristics of subsurface formations are available, effectively synergizing the information from these measurements becomes critical to enhance deep reservoir characterization, determine interwell fluid distribution and ultimately maximize oil recovery. In this study, we develop a featurebased model calibration workflow by combining the power of ensemble methods in data integration and deep learning techniques in feature segmentation. The performance of the developed workflow is demonstrated with a synthetic channelized reservoir model, in which crosswell seismic and electromagnetic (EM) data are jointly inverted.

Enhancing Ensemble Data Assimilation into OneWayCoupled Models with OneStepAheadSmoothing(Quarterly Journal of the Royal Meteorological Society, Wiley, 20200930) [Article]This study investigates the filtering problem with oneway coupled (OWC) statespace systems, for which the joint ensemble Kalman filter (EnKF) is the standard solution. In this approach, the states of the two coupled subsystems are jointly updated with all incoming observations. This enables transferring the information across the subsystems, which should provide coupledstate estimates in better agreement with the observations. The state estimates of the joint EnKF highly depend on the relevance of the joint ensembles’ crosscovariances between the subsystems’ variables. In this work, we propose a new joint EnKF scheme based on the OneStepAhead (OSA) smoothing formulation of the filtering problem for efficient assimilation into OWC systems. The scheme introduces an extra smoothing step for both states subsystems with the future observations, followed by an analysis step for each subsystem state using only its own observation, all within a Bayesian consistent framework. The extra OSAsmoothing step enables to more efficiently exploit the observations, to enhance the representativeness of the EnKF covariances, and to mitigate for reported inconsistencies in the joint EnKF analysis step.We demonstrate the relevance of the proposed approach by presenting and analyzing results of various numerical experiments conducted with a OWC Lorenz96 model.

An efficient multigridDEIM semireducedorder model for simulation of singlephase compressible flow in porous media(Petroleum Science, Springer Science and Business Media LLC, 20200928) [Article]In this paper, an efficient multigridDEIM semireducedorder model is developed to accelerate the simulation of unsteady singlephase compressible flow in porous media. The cornerstone of the proposed model is that the full approximate storage multigrid method is used to accelerate the solution of flow equation in original fullorder space, and the discrete empirical interpolation method (DEIM) is applied to speed up the solution of Peng–Robinson equation of state in reducedorder subspace. The multigridDEIM semireducedorder model combines the computation both in fullorder space and in reducedorder subspace, which not only preserves good prediction accuracy of fullorder model, but also gains dramatic computational acceleration by multigrid and DEIM. Numerical performances including accuracy and acceleration of the proposed model are carefully evaluated by comparing with that of the standard semiimplicit method. In addition, the selection of interpolation points for constructing the lowdimensional subspace for solving the Peng–Robinson equation of state is demonstrated and carried out in detail. Comparison results indicate that the multigridDEIM semireducedorder model can speed up the simulation substantially at the same time preserve good computational accuracy with negligible errors. The general acceleration is up to 50–60 times faster than that of standard semiimplicit method in twodimensional simulations, but the average relative errors of numerical results between these two methods only have the order of magnitude 10−4–10−6%.

Rejoinder to the discussion on A highresolution bilevel skewt stochastic generator for assessing Saudi Arabia's wind energy resources(Environmetrics, Wiley, 20200921) [Article]This is the rejoinder of the discussion article: env190145, DOI: 10.1002/env.2628.

Accelerating flash calculations in unconventional reservoirs considering capillary pressure using an optimized deep learning algorithm(Journal of Petroleum Science and Engineering, Elsevier BV, 20200915) [Article]An increasing focus was placed in the past few decades on accelerating flash calculations and a variety of acceleration strategies have been developed to improve its efficiency without serious compromise in accuracy and reliability. Recently, as machine learning becomes a powerful tool to handle complicated and timeconsuming problems, it is increasingly appealing to replace the iterative flash algorithm, due to the strong nonlinearity of flash problem, by a neural network model. In this study, an NVT flash calculation scheme is established with a thermodynamically stable evolution algorithm to generate training and testing data for the proposed deep neural network. With a modified network structure, the deep learning algorithm is optimized by carefully tuning neural network hyperparameters. Numerical tests indicate that the trained model is capable of accurately estimating phase compositions and states for complex reservoir fluids under a wide range of environmental conditions, while the effect of capillary pressure can be captured well. Thermodynamic rules are preserved well through our algorithm, and the trained model can be used for various fluid mixtures, which significantly accelerates flash calculations in unconventional reservoirs.

Seismological Investigations in the Olduvai Basin and Ngorongoro Volcanic Highlands (Western Flank of the North Tanzanian Divergence)(Seismological Research Letters, Seismological Society of America (SSA), 20200915) [Article]Abstract We present data and results of a passive seismic experiment that we operated between June 2016 and May 2018 in the Ngorongoro Conservation Area (northern Tanzania), located on the western side of the eastern branch of the Eastern African Rift (EAR) system. The motivation for this experiment is twofold: (1) investigating the extension of the Olduvai basin, referred to also as the “Cradle of Human Mankind,” as it hosted a variety of paleoenvironments exploited by hominins during their evolution; and (2) studying the link between the fault system in the main EAR and in its western flank. We conduct detailed dataquality analysis of the seismic recordings based upon ambient noise characterization and numerical waveform simulations. Our data set is of good quality, and we observe that local magnitude can be overestimated up to at least 0.23, due to waveamplifications effects occurring at sites with loose sedimentary material. Based on a new but simple approach using power spectral density measurements, we calculate the thickness of sedimentary basins. This allows us to map the bottom of the Olduvai paleolake confirming that its sedimentary record may be at least 200 m deeper than previously inferred from core drilling. We also map the bottom of the Olbalbal depression for the first time. In addition, we present a seismicity map of the Ngorongoro Conservation Area with unprecedented detail. The seismicity depicts the suture zone between the Tanzanian craton and the Mozambique belt and reveals that the fault system in the western flank of the rift merges at depth into a single detachment that joins the Manyara fault on the western side of the main rift valley.