Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “deep”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 397 records · Page 22

Influence of the Kuroshio Intrusion on Deep Flow Intraseasonal Variability in the Northern South China Sea

Abstract Interactions between the open ocean and marginal seas have been suggested to be critical to the redistribution and dissipation of global energy. Here, we propose a mechanism for the upper open ocean influencing the deep flow in marginal seas that hinges on the formation and propagation of topographic Rossby waves (TRWs). Observations and high‐resolution simulations suggest substantial intraseasonal variability with periods of 5–60 days associated with the deep flow over continental slopes of the northern South China Sea (NSCS). These fluctuations generally account for over 40% of the total deep kinetic energy variability and the number can reach 70% over the slopes to the west of the Luzon Strait, southwest of the Dongsha Islands, and northeast of the Xisha Islands. By examining the spatiotemporal features of the fluctuations, we demonstrate that the intraseasonal variability of the deep flow in the NSCS is closely associated with TRWs. Using a recently developed multiscale energetics analysis in combination with a ray tracing model, we show that the energy sources of TRWs can be traced back to the east of the Dongsha Islands, where the Kuroshio intrusion and related eddies energize the TRWs primarily through pressure work. These waves propagate westward across the NSCS and drive the intraseasonal variability of the deep flow over continental slopes. Our findings highlight an energy pathway from the open ocean western boundary current to the abyssal marginal sea that could modulate regional circulation as well as exchanges between major ocean basins.

Quan, Qi↗

Differences in Tropical Rainfall in Aquaplanet Simulations With Resolved or Parameterized Deep Convection

Abstract This study investigates the effects of resolved deep convection on tropical rainfall and its multi‐scale variability. A series of aquaplanet simulations are analyzed using the Model for Prediction Across Scales‐Atmosphere with horizontal cell spacings from 120 to 3 km. The 3‐km experiment uses a novel configuration with 3‐km cell spacing between 20°S and 20°N and 15‐km cell spacing poleward of 30°N/S. A comparison of those experiments shows that resolved deep convection yields a narrower, stronger, and more equatorward intertropical convergence zone, which is supported by stronger nonlinear horizontal momentum advection in the boundary layer. There is also twice as much tropical rainfall variance in the experiment with resolved deep convection than in the experiments with parameterized convection. All experiments show comparable precipitation variance associated with Kelvin waves; however, the experiment with resolved deep convection shows higher precipitation variance associated with westward propagating systems. Resolved deep convection also yields at least two orders of magnitude more frequent heavy rainfall rates (>2 mm hr −1 ) than the experiments with parameterized convection. A comparison of organized precipitation systems demonstrates that tropical convection organizes into linear systems that are associated with stronger and deeper cold pools and upgradient convective momentum fluxes when convection is resolved. In contrast, parameterized convection results in more circular systems, weaker cold pools, and downgradient convective momentum fluxes. These results suggest that simulations with parameterized convection are missing an important feedback loop between the mean state, convective organization, and meridional gradients of moisture and momentum.

Rios‐Berrios, Rosimar↗

Precipitation–Moisture Coupling Over Tropical Oceans: Sequential Roles of Shallow, Deep, and Mesoscale Convective Systems

We report Precipitation over tropical oceans rapidly increases when the environmental column saturation fraction (CSF) increases past a critical value of ~0.7. Past studies suggested that increased stratiform rainfall greatly contributes to the rapid rainfall enhancement. In this study, the sequential roles of non-deep convection, deep convection, and mesoscale convective system (MCS) in precipitation-moisture interactions are examined using 19 years of satellite observations. When CSF is below ~0.5, non-deep convection dominates total rainfall, and predominantly contributes to moistening of the environment. Between the CSF range of 0.5–0.7, transition to deep convective rainfall begins. Meanwhile, MCS contribution to total rain rapidly increases, and the environment is further moistened. MCS becomes the major rainfall type above the critical CSF value (~0.7), with the rapid increase of total rain mostly explained by the rapid increase in MCS rain area. Rainfall reduction at high CSF values is jointly contributed by MCS and non-deep convection.

58 GEOSCIENCES↗

Combining Deep Learning With Physics Based Features in Explosion-Earthquake Discrimination

This paper combines the power of deep-learning with the generalizability of physics-based features, to present an advanced method for seismic discrimination between earthquakes and explosions. The proposed method contains two branches: a deep learning branch operating directly on seismic waveforms or spectrograms, and a second branch operating on physics-based parametric features. These features are high-frequency P/S amplitude ratios and the difference between local magnitude (ML) and coda duration magnitude (MC). The combination achieves better generalization performance when applied to new regions than models that are developed solely with deep learning. Further, we also examined which parts of the waveform data dominate deep learning decisions (i.e., via Grad-CAM). Such visualization provides a window into the black-box nature of the machine-learning models and offers new insight into how the deep learning derived models use data to make decisions.

58 GEOSCIENCES↗

High-pressure orthorhombic ferromagnesite as a potential deep-mantle carbon carrier

In this study, knowledge of the physical and chemical properties of candidate deep-carbon carriers such as ferromagnesite [(Mg,Fe)CO 3 ] at high pressure and temperature of the deep mantle is necessary for our understanding of deep-carbon storage as well as the global carbon cycle of the planet. Previous studies have reported very different scenarios for the (Mg,Fe)CO 3 system at deep-mantle conditions including the chemical dissociation to (Mg,Fe)O+CO 2 , the occurrence of the tetrahedrally-coordinated carbonates based on CO 4 structural units, and various high-pressure phase transitions. Here we have studied the phase stability and compressional behavior of (Mg,Fe)CO 3 carbonates up to relevant lower-mantle conditions of approximately 120 GPa and 2400 K. Our experimental results show that the rhombohedral siderite (Phase I) transforms to an orthorhombic phase (Phase II with Pmm2 space group) at approximately 50 GPa and 1400 K. The structural transition is likely driven by the spin transition of iron accompanied by a volume collapse in the Fe-rich (Mg,Fe)CO 3 phases; the spin transition stabilizes the high-pressure phase II at much lower pressure conditions than its Mg-rich counterpart. It is conceivable that the low-spin ferromagnesite phase II becomes a major deep-carbon carrier at the deeper parts of the lower mantle below 1900 km in depth.

36 MATERIALS SCIENCE↗

Inverse deep learning methods and benchmarks for artificial electromagnetic material design

In this work we investigate the use of deep inverse models (DIMs) for designing artificial electromagnetic materials (AEMs) – such as metamaterials, photonic crystals, and plasmonics – to achieve some desired scattering properties (e.g., transmission or reflection spectrum). DIMs are deep neural networks (i.e., deep learning models) that are specially-designed to solve ill-posed inverse problems. There has recently been tremendous growth in the use of DIMs for solving AEM design problems however there has been little comparison of these approaches to examine their absolute and relative performance capabilities. In this work we compare eight state-of-the-art DIMs on three unique AEM design problems, including two models that are novel to the AEM community. Our results indicate that DIMs can rapidly produce accurate designs to achieve a custom desired scattering on all three problems. Although no single model always performs best, the Neural-Adjoint approach achieves the best overall performance across all problem settings. As a final contribution we show that not all AEM design problems are ill-posed, and in such cases a conventional deep neural network can perform better than DIMs. We recommend that a deep neural network is always employed as a simple baseline approach when addressing AEM design problems. Furthermore, we publish python code for our AEM simulators and our DIMs to enable easy replication of our results, and benchmarking of new DIMs by the AEM community.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Dark Energy Survey Deep Field photometric redshift performance and training incompleteness assessment

Context. The determination of accurate photometric redshifts (photo-zs) in large imaging galaxy surveys is key for cosmological studies. One of the most common approaches are machine learning techniques. These methods require a spectroscopic or reference sample to train the algorithms. Attention has to be paid to the quality and properties of these samples since they are key factors in the estimation of reliable photo-zs. Aims. The goal of this work is to calculate the photo-zs for the Y3 DES Deep Fields catalogue using the DNF machine learning algorithm. Moreover, we want to develop techniques to assess the incompleteness of the training sample and metrics to study how incompleteness affects the quality of photometric redshifts. Finally, we are interested in comparing the performance obtained with respect to the EAzY template fitting approach on Y3 DES Deep Fields catalogue. Methods. We have emulated -- at brighter magnitude -- the training incompleteness with a spectroscopic sample whose redshifts are known to have a measurable view of the problem. We have used a principal component analysis to graphically assess incompleteness and to relate it with the performance parameters provided by DNF. Finally, we have applied the results about the incompleteness to the photo-z computation on Y3 DES Deep Fields with DNF and estimated its performance. Results. The photo-zs for the galaxies on DES Deep Fields have been computed with the DNF algorithm and added to the Y3 DES Deep Fields catalogue. They are available at https://des.ncsa.illinois.edu/releases/y3a2/Y3deepfields. Some techniques have been developed to evaluate the performance in the absence of "true" redshift and to assess completeness. We have studied... (Partial abstract)

79 ASTRONOMY AND ASTROPHYSICS↗

Atomic force microscopy deep trench and sidewall imaging with an optical fiber probe

We report a method to measure critical dimensions of micro- and nanostructures using the atomic force microscope (AFM) with an optical fiber probe (OFP). This method is capable of scanning narrow and deep trenches due to the long and thin OFP tip, as well as imaging of steep sidewalls with unique profiling possibilities by laterally tilting the OFP without any modifications of the optical lever. A switch control scheme is developed to measure the sidewall angle by flexibly transferring feedback control between the Z- and Y-axis, for a serial scan of the horizontal surface (raster scan on XY-plane) and sidewall (raster scan on the YZ-plane), respectively. In experiments, a deep trench with tapered walls (243.5 μm deep) and a microhole (about 14.9 μm deep) have been imaged with the orthogonally aligned OFP, as well as a silicon sidewall (fabricated by deep reactive ion etching) has been characterized with the tilted OFP. Moreover, the sidewall angle of TGZ3 (AFM calibration grating) was accurately measured using the switchable scan method.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Capacitance behavior of InAlN Schottky diodes in presence of large concentrations of shallow and deep states related to oxygen

The capacitance-voltage-temperature characteristics of nonintentionally doped In{sub 0.16}Al{sub 0.84 }N/n{sup +}-GaN Schottky diodes were measured at 1 MHz and in the 90–400 K range. They are discussed in the framework of existing theories, which properly treat the Poisson's equation, especially near the edge of the space-charge region, the so-called transition region. The concentration of a shallow donor and of a deep DX-like center, previously reported, is properly determined. The key parameter to discuss the temperature dependence of the capacitance is the ratio between the frequency of the small ac modulating signal and the temperature-dependent emission rate associated to each level. The capacitance-voltage C-V{sub a} curves were successfully fitted using a three parameters expression over the full range of temperatures. The concentration of both shallow and deep levels exceeds a few 10{sup 18} cm{sup −3}. Based on secondary ion mass spectrometry profiling, we assign both levels to the dominant oxygen impurity. This result supports our previous assignment of the shallow donor to a substitutional oxygen atom on a nitrogen site and the deep state to an O-related DX center, naturally explaining its high concentration. The sluggish kinetics at low temperatures, associated to the large concentration of deep levels located near the transition region, is illustrated by hysteresis loops in the C-V{sub a} curves below 270 K. Furthermore, the contribution of free carriers to the capacitance is revealed below 150 K, when both shallow and deep donors cannot respond anymore due to an emission rate lower than the 1 MHz modulating frequency. Finally, the presence of a highly doped thin surface barrier, as already reported in other III-nitrides, finds further support.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

High temperature annealing effects on deep-level defects in a high purity semi-insulating 4H-SiC substrate

Effects of high-temperature annealing on deep-level defects in a high-purity semi-insulating 4H silicon carbide substrate have been studied by employing current-voltage, capacitance-voltage, junction spectroscopy, and chemical impurity analysis measurements. Secondary ion mass spectrometry data reveal that the substrate contains boron with concentration in the mid 10{sup 15 }cm{sup −3} range, while other impurities including nitrogen, aluminum, titanium, vanadium and chromium are below their detection limits (typically ∼10{sup 14 }cm{sup −3}). Schottky barrier diodes fabricated on substrates annealed at 1400–1700 °C exhibit metal/p-type semiconductor behavior with a current rectification of up to 8 orders of magnitude at bias voltages of ±3 V. With increasing annealing temperature, the series resistance of the Schottky barrier diodes decreases, and the net acceptor concentration in the substrates increases approaching the chemical boron content. Admittance spectroscopy results unveil the presence of shallow boron acceptors and deep-level defects with levels in lower half of the bandgap. After the 1400 °C annealing, the boron acceptor still remains strongly compensated at room temperature by deep donor-like levels located close to mid-gap. However, the latter decrease in concentration with increasing annealing temperature and after 1700 °C, the boron acceptor is essentially uncompensated. Hence, the deep donors are decisive for the semi-insulating properties of the substrates, and their thermal evolution limits the thermal budget for device processing. The origin of the deep donors is not well-established, but substantial evidence supporting an assignment to carbon vacancies is presented.

36 MATERIALS SCIENCE↗

Evolution of AlGaN deep level defects as a function of alloying and compositional grading and resultant impact on electrical conductivity

AlGaN:Si epilayers with uniform Al compositions of 60%, 70%, 80%, and 90% were grown by metal-organic vapor phase epitaxy along with a compositionally graded, unintentionally doped (UID) AlGaN epilayer with the Al composition varying linearly between 80% and 100%. The resistivity of AlGaN:Si with a uniform composition increased significantly for the Al content of 80% and greater, whereas the graded UID-AlGaN film exhibited resistivity equivalent to 60% and 70% AlGaN:Si owing to polarization-induced doping. Deep level defect studies of both types of AlGaN epilayers were performed to determine why the electronic properties of uniform-composition AlGaN:Si degraded with increased Al content, while the electronic properties of graded UID-AlGaN did not. The deep level density of uniform-composition AlGaN:Si increased monotonically and significantly with the Al mole fraction. Conversely, graded-UID AlGaN had the lowest deep level density of all the epilayers despite containing the highest Al composition. These findings indicate that Si doping is an impetus for point defect incorporation in AlGaN that becomes stronger with the increasing Al content. However, the increase in deep level density with the Al content in uniform-composition AlGaN:Si was small compared to the increase in resistivity. This implies that the primary cause for increasing resistivity in AlGaN:Si with the increasing Al mole fraction is not compensation by deep levels but rather increasing activation energy for the Si dopant. As a result, the graded UID-AlGaN films maintained low resistivity because they do not rely on thermal ionization of Si dopants.

36 MATERIALS SCIENCE↗

THE DRAO PLANCK DEEP FIELDS: THE POLARIZATION PROPERTIES OF RADIO GALAXIES AT 1.4 GHz

We present results of deep polarization imaging at 1.4 GHz with the Dominion Radio Astrophysical Observatory as part of the DRAO Planck Deep Fields project. This deep extragalactic field covers 15.16 deg{sup 2} centered at {alpha}{sub 2000} = 16{sup h}14{sup m} and {delta}{sub 2000} = 54{sup 0}56', has an angular resolution of 42'' x 62'' at the field center, and reaches a sensitivity of 55 {mu}Jy beam{sup -1} in Stokes I and 45 {mu}Jy beam{sup -1} in Stokes Q and U. We detect 958 radio sources in Stokes I of which 136 are detected in polarization. We present the Euclidean-normalized polarized differential source counts down to 400 {mu}Jy. These counts indicate that sources have a higher degree of fractional polarization at fainter Stokes I flux density levels than for brighter sources, confirming an earlier result. We find that the majority of our polarized sources are steep-spectrum objects with a mean spectral index of -0.77, and there is no correlation between fractional polarization and spectral index. We also matched deep field sources to counterparts in the Faint Images of the Radio Sky at Twenty Centimeters catalog. Of the polarized sources, 77% show structure at the arcsecond scale whereas only 38% of the sources with no detectable polarization show such structure. The median fractional polarization for resolved sources is 6.8%, while it is 4.4% for compact objects. The polarized radio sources in our deep field are predominantly those sources which are resolved and show the highest degrees of fractional polarization, indicating that the lobe dominated structure may be the source of the highly polarized sources. These resolved radio galaxies dominate the polarized source counts at P{sub 0}={radical}(Q{sup 2} + U{sup 2})<3 mJy.

79 ASTRONOMY AND ASTROPHYSICS↗

ON THE VARIATION OF ZONAL GRAVITY COEFFICIENTS OF A GIANT PLANET CAUSED BY ITS DEEP ZONAL FLOWS

Rapidly rotating giant planets are usually marked by the existence of strong zonal flows at the cloud level. If the zonal flow is sufficiently deep and strong, it can produce hydrostatic-related gravitational anomalies through distortion of the planet's shape. This paper determines the zonal gravity coefficients, J{sub 2n}, n = 1, 2, 3, ..., via an analytical method taking into account rotation-induced shape changes by assuming that a planet has an effective uniform density and that the zonal flows arise from deep convection and extend along cylinders parallel to the rotation axis. Two different but related hydrostatic models are considered. When a giant planet is in rigid-body rotation, the exact solution of the problem using oblate spheroidal coordinates is derived, allowing us to compute the value of its zonal gravity coefficients J-bar{sub 2n}, n=1,2,3,..., without making any approximation. When the deep zonal flow is sufficiently strong, we develop a general perturbation theory for estimating the variation of the zonal gravity coefficients, {Delta}J{sub 2n}=J{sub 2n}-J-bar{sub 2n}, n=1,2,3,..., caused by the effect of the deep zonal flows for an arbitrarily rapidly rotating planet. Applying the general theory to Jupiter, we find that the deep zonal flow could contribute up to 0.3% of the J{sub 2} coefficient and 0.7% of J{sub 4}. It is also found that the shape-driven harmonics at the 10th zonal gravity coefficient become dominant, i.e., {Delta}J{sub 2n}>=J-bar{sub 2n} for n {>=} 5.

79 ASTRONOMY AND ASTROPHYSICS↗

EMPIRICAL PREDICTIONS FOR (SUB-)MILLIMETER LINE AND CONTINUUM DEEP FIELDS

Modern (sub-)millimeter/radio interferometers such as ALMA, JVLA, and the PdBI successor NOEMA will enable us to measure the dust and molecular gas emission from galaxies that have luminosities lower than the Milky Way, out to high redshifts and with unprecedented spatial resolution and sensitivity. This will provide new constraints on the star formation properties and gas reservoir in galaxies throughout cosmic times through dedicated deep field campaigns targeting the CO/[C II] lines and dust continuum emission in the (sub-)millimeter regime. In this paper, we present empirical predictions for such line and continuum deep fields. We base these predictions on the deepest available optical/near-infrared Advanced Camera for Surveys and NICMOS data on the Hubble Ultra Deep Field (over an area of about 12 arcmin{sup 2}). Using a physically motivated spectral energy distribution model, we fit the observed optical/near-infrared emission of 13,099 galaxies with redshifts up to z = 5, and obtain median-likelihood estimates of their stellar mass, star formation rate, dust attenuation, and dust luminosity. We combine the attenuated stellar spectra with a library of infrared emission models spanning a wide range of dust temperatures to derive statistical constraints on the dust emission in the infrared and (sub-)millimeter which are consistent with the observed optical/near-infrared emission in terms of energy balance. This allows us to estimate, for each galaxy, the (sub-)millimeter continuum flux densities in several ALMA, PdBI/NOEMA, and JVLA bands. As a consistency check, we verify that the 850 {mu}m number counts and extragalactic background light derived using our predictions are consistent with previous observations. Using empirical relations between the observed CO/[C II] line luminosities and the infrared luminosity of star-forming galaxies, we infer the luminosity of the CO(1-0) and [C II] lines from the estimated infrared luminosity of each galaxy in our sample. We then predict the luminosities of higher CO transition lines CO(2-1) to CO(7-6) based on two extreme gas excitation scenarios: quiescent (Milky Way) and starburst (M82). We use our predictions to discuss possible deep field strategies with ALMA. The predictions presented in this study will serve as a direct benchmark for future deep field campaigns in the (sub-)millimeter regime.

79 ASTRONOMY AND ASTROPHYSICS↗

AEGIS-X: Deep Chandra imaging of the central Groth strip

We present the results of deep Chandra imaging of the central region of the Extended Groth Strip, the AEGIS-X Deep (AEGIS-XD) survey. When combined with previous Chandra observations of a wider area of the strip, AEGIS-X Wide (AEGIS-XW), these provide data to a nominal exposure depth of 800 ks in the three central ACIS-I fields, a region of approximately 0.29 deg{sup 2}. This is currently the third deepest X-ray survey in existence; a factor ∼2−3 shallower than the Chandra Deep Fields (CDFs), but over an area ∼3 times greater than each CDF. We present a catalog of 937 point sources detected in the deep Chandra observations, along with identifications of our X-ray sources from deep ground-based, Spitzer, GALEX, and Hubble Space Telescope imaging. Using a likelihood ratio analysis, we associate multiband counterparts for 929/937 of our X-ray sources, with an estimated 95% reliability, making the identification completeness approximately 94% in a statistical sense. Reliable spectroscopic redshifts for 353 of our X-ray sources are available predominantly from Keck (DEEP2/3) and MMT Hectospec, so the current spectroscopic completeness is ∼38%. For the remainder of the X-ray sources, we compute photometric redshifts based on multiband photometry in up to 35 bands from the UV to mid-IR. Particular attention is given to the fact that the vast majority the X-ray sources are active galactic nuclei and require hybrid templates. Our photometric redshifts have mean accuracy of σ=0.04 and an outlier fraction of approximately 5%, reaching σ=0.03 with less than 4% outliers in the area covered by CANDELS . The X-ray, multiwavelength photometry, and redshift catalogs are made publicly available.

79 ASTRONOMY AND ASTROPHYSICS↗

A deep dilated convolutional residual network for predicting interchain contacts of protein homodimers

Abstract Motivation Deep learning has revolutionized protein tertiary structure prediction recently. The cutting-edge deep learning methods such as AlphaFold can predict high-accuracy tertiary structures for most individual protein chains. However, the accuracy of predicting quaternary structures of protein complexes consisting of multiple chains is still relatively low due to lack of advanced deep learning methods in the field. Because interchain residue–residue contacts can be used as distance restraints to guide quaternary structure modeling, here we develop a deep dilated convolutional residual network method (DRCon) to predict interchain residue–residue contacts in homodimers from residue–residue co-evolutionary signals derived from multiple sequence alignments of monomers, intrachain residue–residue contacts of monomers extracted from true/predicted tertiary structures or predicted by deep learning, and other sequence and structural features. Results Tested on three homodimer test datasets (Homo_std dataset, DeepHomo dataset and CASP-CAPRI dataset), the precision of DRCon for top L/5 interchain contact predictions (L: length of monomer in a homodimer) is 43.46%, 47.10% and 33.50% respectively at 6 Å contact threshold, which is substantially better than DeepHomo and DNCON2_inter and similar to Glinter. Moreover, our experiments demonstrate that using predicted tertiary structure or intrachain contacts of monomers in the unbound state as input, DRCon still performs well, even though its accuracy is lower than using true tertiary structures in the bound state are used as input. Finally, our case study shows that good interchain contact predictions can be used to build high-accuracy quaternary structure models of homodimers. Availability and implementation The source code of DRCon is available at https://github.com/jianlin-cheng/DRCon. The datasets are available at https://zenodo.org/record/5998532#.YgF70vXMKsB. Supplementary information Supplementary data are available at Bioinformatics online.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Optimizing Kernel Machines Using Deep Learning

Building highly non-linear and non-parametric models is central to several state-of-the-art machine learning systems. Kernel methods form an important class of techniques that induce a reproducing kernel Hilbert space (RKHS) for inferring non-linear models through the construction of similarity functions from data. These methods are particularly preferred in cases where the training data sizes are limited and when prior knowledge of the data similarities is available. Despite their usefulness, they are limited by the computational complexity and their inability to support end-to-end learning with a task specific objective. On the other hand, deep neural networks have become the de facto solution for end-to-end inference in several learning paradigms. In this article, we explore the idea of using deep architectures to perform kernel machine optimization, for both computational efficiency and end-to-end inferencing. To this end, we develop the DKMO (Deep Kernel Machine Optimization) framework, that creates an ensemble of dense embeddings using Nystrõm kernel approximations and utilizes deep learning to generate task-specific representations through the fusion of the embeddings. Intuitively, the filters of the network are trained to fuse information from an ensemble of linear subspaces in the RKHS. Furthermore, we introduce the kernel dropout regularization to enable improved training convergence. Finally, we extend this framework to the multiple kernel case, by coupling a global fusion layer with pre-trained deep kernel machines for each of the constituent kernels. Using case studies with limited training data, and lack of explicit feature sources, we demonstrate the effectiveness of our framework over conventional model inferencing techniques.

97 MATHEMATICS AND COMPUTING↗

Deep learning multiphysics network for imaging CO 2 saturation and estimating uncertainty in geological carbon storage

Multiphysics inversion exploits different types of geophysical data that often complement each other and aims to improve overall imaging resolution and reduce uncertainties in geophysical interpretation. Despite the advantages, traditional multiphysics inversion is challenging because it requires a large amount of computational time and intensive human interactions for preprocessing data and finding trade-off parameters. These issues make it nearly impossible for traditional multiphysics inversion to be applied as a real-time monitoring tool for geological carbon storage. In this paper, we present a deep learning (DL) multiphysics network for imaging CO 2 saturation in real time. The multiphysics network consists of three encoders for analysing seismic, electromagnetic and gravity data and shares one decoder for combining imaging capabilities of the different geophysical data for better predicting CO 2 saturation. The network is trained on pairs of CO 2 label models and multiphysics data so that it can directly image CO 2 saturation. Here we use the bootstrap aggregating method to enhance the imaging accuracy and estimate uncertainties associated with CO 2 saturation images. Using realistic CO 2 label models and multiphysics data derived from the Kimberlina CO 2 storage model, we evaluate the performance of the deep learning multiphysics network and compare its imaging results to those from the deep learning single-physics networks. Our modelling experiments show that the deep learning multiphysics network for seismic, electromagnetic, and gravity data not only improves the imaging accuracy but also reduces uncertainties associated with CO 2 saturation images. Our results also suggest that the deep learning multiphysics network for the non-seismic data (i.e., electromagnetic and gravity) can be used as an effective low-cost monitoring tool in between regular seismic monitoring.

58 GEOSCIENCES↗