Results 1  10
of
13
CWP651 Full waveform inversion with imageguided gradient
"... Figure 1. Change of data misfit functions vs. iterations in full waveform inversion and imageguided full waveform inversion. The objective of seismic full waveform inversion (FWI) is to estimate a model of the subsurface that minimizes the difference between recorded seismic data and synthetic data ..."
Abstract

Cited by 4 (3 self)
 Add to MetaCart
(Show Context)
Figure 1. Change of data misfit functions vs. iterations in full waveform inversion and imageguided full waveform inversion. The objective of seismic full waveform inversion (FWI) is to estimate a model of the subsurface that minimizes the difference between recorded seismic data and synthetic data simulated in that model. Although FWI can yield accurate and highresolution models, multiple problems have prevented widespread application of this technique in practice. First, FWI is computationally intensive, in part because it typically requires many iterations of costly gradientdescent calculations to converge to a solution model. Second, FWI often converges to spurious local minima in the data misfit function of the difference between recorded and synthetic data. Third, FWI is an underdetermined inverse problem with many solutions, most of which may make no geological sense. These problems are related to a typically large number of model parameters and to the absence of low frequencies in recorded data. FWI with an imageguided gradient mitigates these problems by reducing the number of parameters in the subsurface model. We represent the subsurface model with a sparse set of values, and from these values, we use imageguided interpolation (IGI) to compute finely and uniformlysampled gradients of the data misfit function in FWI. Because the interpolation is guided by seismic images, gradients computed in this way conform to geologic structures and subsequently yield models that also agree with subsurface structures. Because of sparse parametrization in the model space, IGI creates models that are more blocky than finelysampled models, and this blockiness from the model space mitigates the absence of low frequencies in recorded data. A smaller number of parameters to invert also reduces the number of iterations required to converge to a solution model. Tests with a synthetic model and data demonstrate these improvements.
CWP654 Painting seismic images in 3D
"... Figure 1. A seismic image painted using a 3D digital paintbrush that conforms to features in the image Seismic interpretation today includes picking seismic horizon surfaces or, more generally, the boundary between geologic bodies. A more efficient and useful approach may be to directly interpret th ..."
Abstract

Cited by 1 (0 self)
 Add to MetaCart
Figure 1. A seismic image painted using a 3D digital paintbrush that conforms to features in the image Seismic interpretation today includes picking seismic horizon surfaces or, more generally, the boundary between geologic bodies. A more efficient and useful approach may be to directly interpret those geologic bodies as 3D volumes. We do this by painting voxels in 3D seismic images of subsurface geology. In our painting method, a human interpreter controls the maximum size of a digital 3D paintbrush, and as the interpreter interactively moves the brush, features in the 3D seismic image automatically control its shape, orientation and size. Key words: 3D painting seismic interpretation 1
CWP656 Imageguided 3D interpolation of borehole data
"... (b) of those measured velocities. A blended neighbor method for imageguided interpolation enables resampling of borehole data onto a uniform 3D sampling grid, without picking horizons and without flattening seismic images. Borehole measurements gridded in this way become new 3D images of subsurface ..."
Abstract

Cited by 1 (1 self)
 Add to MetaCart
(Show Context)
(b) of those measured velocities. A blended neighbor method for imageguided interpolation enables resampling of borehole data onto a uniform 3D sampling grid, without picking horizons and without flattening seismic images. Borehole measurements gridded in this way become new 3D images of subsurface properties. Property values conform to geologic layers and faults apparent in the seismic image that guided the interpolation. The freely available Teapot Dome data set, which includes a 3D seismic image, horizons picked from that image, and numerous well logs, provides an ideal demonstration of imageguided interpolation of borehole data. In this example, seismic horizons picked by others coincide with thin layers apparent in the new 3D images of interpolated borehole data, even though the horizons were not used in the interpolation process. Key words: seismic image well logs interpolation interpretation 1
CWP722 Unfaulting and unfolding 3D seismic images
"... a) b) c) Figure 1. A seismic image (a) is first unfaulted (b) and then flattened (c) using fault throw vectors and flattening shift vectors computed automatically. One limitation of automatic interpretation methods such as seismic image flattening is their inability to handle geologic faults. To ad ..."
Abstract

Cited by 1 (1 self)
 Add to MetaCart
a) b) c) Figure 1. A seismic image (a) is first unfaulted (b) and then flattened (c) using fault throw vectors and flattening shift vectors computed automatically. One limitation of automatic interpretation methods such as seismic image flattening is their inability to handle geologic faults. To address this limitation, we propose to combine a method for automatic image unfaulting with seismic image flattening. First, using fault surfaces and fault throw vectors estimated from an image, we interpolate throw vectors to produce a throw vector field, which we use to unfault the image. Then, we flatten the unfaulted image to obtain a new image in which sedimentary layering is horizontal and also aligned across faults. From this flattened unfaulted image, we can automatically extract geologic horizons. Key words: seismic image unfaulting flattening interpretation interpolation 1
CWP678 Imageguided full waveform inversion
"... Both inversions use highly smoothed Marmousi II velocity as the initial model; a 15Hz Ricker wavelet is used as the source for the 11 shots in the inversion. Multiple problems, including high computational cost, spurious local minima, and solutions with no geologic sense, have prevented widespread a ..."
Abstract
 Add to MetaCart
(Show Context)
Both inversions use highly smoothed Marmousi II velocity as the initial model; a 15Hz Ricker wavelet is used as the source for the 11 shots in the inversion. Multiple problems, including high computational cost, spurious local minima, and solutions with no geologic sense, have prevented widespread application of full waveform inversion (FWI), especially FWI of seismic reflections. These problems are fundamentally related to a large number of model parameters and to the absence of low frequencies in recorded seismograms. Instead of inverting for all the parameters in a dense model, imageguided full waveform inversion inverts for a sparse model space that contains far fewer parameters. We represent a model with a sparse set of values, and from these values, we use imageguided interpolation (IGI) and its adjoint operator to compute finely and uniformlysampled models that can fit recorded data in FWI. Because of this sparse representation, imageguided FWI updates more blocky models, and this blockiness in the model space mitigates the absence of low frequencies in recorded data. Moreover, IGI honors imaged structures, so imageguided FWI built in this way yields models that are geologically sensible. Key words: imageguided, full waveform inversion, reduced model space 1
CWP679 A projected Hessian for full waveform inversion
"... Figure 1. Update directions for one iteration of the conjugate gradient method (a), the imageguided conjugate gradient method (b), and a quasiNewton method with application of the inverse projected Hessian (c). A Hessian matrix in full waveform inversion (FWI) is difficult to compute directly beca ..."
Abstract
 Add to MetaCart
(Show Context)
Figure 1. Update directions for one iteration of the conjugate gradient method (a), the imageguided conjugate gradient method (b), and a quasiNewton method with application of the inverse projected Hessian (c). A Hessian matrix in full waveform inversion (FWI) is difficult to compute directly because of high computational cost and an especially large memory requirement. Therefore, Newtonlike methods are rarely feasible in realistic largesize FWI problems. We modify the BFGS method to use a projected Hessian matrix that reduces both the computational cost and memory required, thereby making a quasiNewton solution to FWI feasible. Key words: projected Hessian matrix, BFGS method 1
cwp697 Tensorguided interpolation of subducting slab depths
"... Figure 1. Scattered depth samples (a) from the subducting slab in South America with blended neighbor interpolations of depths accounting for (b) the curvature of the earth’s surface, and (c) both that curvature and estimated slab strikes. White ellipses represent models of spatial correlation. The ..."
Abstract
 Add to MetaCart
(Show Context)
Figure 1. Scattered depth samples (a) from the subducting slab in South America with blended neighbor interpolations of depths accounting for (b) the curvature of the earth’s surface, and (c) both that curvature and estimated slab strikes. White ellipses represent models of spatial correlation. The solid lines represent the west coast of South America. Earthquakes and seismic surveys provide estimates of depths of subducting slabs, but only at scattered locations. To construct a useful three dimensional model of slab geometry we must interpolate slab depths at uniformly sampled locations on the earth’s surface, and the interpolation method should account for the curvature of that surface. In addition to estimates of depths from earthquake locations, focal mechanisms of subduction zone earthquakes also provide estimates of the strikes and dips of the subducting slab on which they occur. We use estimated strikes to construct a metric tensor field that guides a blended neighbor interpolation of estimated depths. Our interpolation serves as an example of using one set of scattered data to infer the spatial correlation of another, while accounting for the acquisition of those data on a nonplanar surface.
CWP696 Tensorguided interpolation on nonplanar surfaces
"... a) b) c) Figure 1. 65 monthly averages of atmospheric CO2 measurements acquired at scattered locations on the earth’s surface (a), and tensorguided interpolation of those measurements in a 2D parametric space (b) to obtain interpolated CO2 concentrations everywhere on that surface (c). In blendedn ..."
Abstract
 Add to MetaCart
(Show Context)
a) b) c) Figure 1. 65 monthly averages of atmospheric CO2 measurements acquired at scattered locations on the earth’s surface (a), and tensorguided interpolation of those measurements in a 2D parametric space (b) to obtain interpolated CO2 concentrations everywhere on that surface (c). In blendedneighbor interpolation of scattered data, a tensor field represents a model of spatial correlation that is both anisotropic and spatially varying. In effect, this tensor field defines a nonEuclidean metric, a measure of distance that varies with direction and location. The tensors may be derived from secondary data, such as images. Alternatively, when the primary data to be interpolated are measured on a nonplanar surface, the tensors may be derived from surface geometry, and the nonEuclidean measure of distance is simply geodesic. Interpolation of geophysical data acquired on a nonplanar surface should be consistent with tensor fields derived from both surface geometry and any secondary data. Key words: tensor interpolation parametric surface 1