N. Gutierrez

Lossy Data Compression with DCT Transforms

In todays computers, disks are the slowest among the performance bottlenecks presented by a computer. In order to overcome their limitations, researchers/developers must re-engineer their codes to save less often data to disk and to reduce the amount of saved data. This paper focuses on overcoming the limitations imposed by the local disk in terms of transfer time and total size of the resulting input/output (I/O) operations. To do this, we have developed a variation of a well-known family of image compression algorithms (Discrete Cosine Transform), that have already proved their properties in terms of image bandwidth compression and precision. By using this compression scheme, we are able to perform the simulation without having to use additional techniques as check-pointing, or random boundaries. We show that, with a well-designed compression algorithm, the time to the result can be dramatically reduced, as well as the disk required to process it. Finally, we show the comparison of a raw industrial, synthetic shot compared to that processed with our scheme, showing that the differences among them are affordable whilst the compression ratio obtained is considerable.

Electromagnetic Modeling Using a Massively Parallel Software Framework

Parallel computer systems for geophysical electromagnetic modeling and inversion problems have been developed rapidly in the past decade addressing the needs of the industrial applications. In this work we present electromagnetics in BSIT, a software framework developed for massively parallel distributed environments. Its flexible structure allows to combine different physical problems with various solution methods and computer architectures. Inversion and forward modeling can be performed on independent grids with a proper interpolation scheme. For the EM modeling, kernels for isotropic and transversely isotropic media have been developed. The simulations show high efficiency and scalability of the code. Joint inversion of EM and other types of geophysical data could be performed in the future.

Efficient Lossy Compression for Seismic Processing

Most advanced imaging tools available nowadays need massive computing capabilities. Reverse time migration (RTM) and full waveform inversion (FWI) are among the most sophisticated tools for seismic imaging. As both tools are based upon the simulation of the complete seismic wavefield, large computing platforms are required in order to use such imaging tools for large 3D surveys. One of the most important computational issues is related with the amount of data that need to be transfer to/from disk, considering that I/O is usually orders of magnitude slower than computational resources such as memory or processors. Hence, reducing the pressure over the I/O is a key topic for any seismic system in order to obtain good performance. In this work we present a compression method based on a re-quantization algorithm, that use less bits to represent each value than those needed in the original dataset.

Elastic Mimetic Finite-differences in the Presence of Topography

Finite-differences (FD) are the most popular schemes for seismic modelling. For elastic problems, staggered-grid FD offers many advantages in terms of accuracy and efficiency. However, FD methods typically struggle at modelling seismic waves in the presence of a non-flat topography. This is due to two main issues: First of all, the free-surface boundary condition is badly represented with classical differentiating operators and, additionally, the topography is poorly represented by regular Cartesian grids. We introduce a new FD scheme which combines mimetic finite-difference operators with a deformed-grid approach to successfully model elastic waves in 3D scenarios including topography. Our results show that our scheme produces precise results with few points per wavelength, which translates in low computational requirements for large 3D simulations, and compares well with a discontinuous Galerkin method.

Acceleration strategies for elastic full waveform inversion workflows in 2D and 3D

Full waveform inversion (FWI) is one of the most challenging procedures to obtain quantitative information of the subsurface. For elastic inversions, when both compressional and shear velocities have to be inverted, the algorithmic issue becomes also a computational challenge due to the high cost related to modelling elastic rather than acoustic waves. This shortcoming has been moderately mitigated by using high-performance computing to accelerate 3D elastic FWI kernels. Nevertheless, there is room in the FWI workflows for obtaining large speedups at the cost of proper grid pre-processing and data decimation techniques. In the present work, we show how by making full use of frequency-adapted grids, composite shot lists and a novel dynamic offset control strategy, we can reduce by several orders of magnitude the compute time while improving the convergence of the method in the studied cases, regardless of the forward and adjoint compute kernels used.

3D Elastic Full Waveform Inversion - On Land Study case

Full Waveform Inversion is one of the most advanced processing methods that is recently reaching a mature state after years of solving theoretical and technical issues such as the non-uniqueness of the solution and harnessing the huge computational power required by realistic scenarios. In this work, we present the application of this method to a 3D on-land dataset acquired to characterize the shallow subsurface. The current study explores the possibility to apply elastic isotropic Full Waveform Inversion using only the vertical component of the recorded seismograms. One of the main challenges in this case study remains the costly 3D modeling that includes topography and free surface effects. Nevertheless, the resulting models provide a higher resolution of the subsurface structures than starting models, and show a good correlation with the available borehole measurements.

Elastic Full Waveform Inversion (FWI) of reflection data with a phase misfit function

Full Waveform Inversion of elastic dataset is challenging due to the complexity introduced by free-surface effects or P-S wave conversions among others. In this context, large offsets are preferred for inversion because they favor transmission modes which are more linearly related to P-wave velocity. In this paper, we present an original approach which allows to dynamically select the near offset at each frequency. We illustrate this approach with the inversion of a dataset without density. In order to deal with a more realistic scenario, we next present the inversion with density effects included into the modeling. As inverting density is known to be a hard task, we choose to not invert it. This approach leads to the use of a phase misfit function, which is more connected to the kinematics of the problem than the classic \(L^2\) norm.

Developing Full Waveform Inversion Using HPC Frameworks: BSIT

Full Waveform Inversion (FWI) is a geophysical tool that is designed to improve subsurface models by directly comparing, by means of a misfit function, synthetic traces obtained using an initial model with the real trace recorded experimentally. However, the high computational cost of the forward modeling combined with the huge size of the data, make FWI an extremely complex HPC problem. Nevertheless, its enhanced resolution compared to other seismic inversion methods is starting to make FWI an attractive subsurface imaging tool for both industry and academy. An FWI system workflow consists in processing the data (e.g filtering, windowing), launching gradi- ent computation simulations for each shot, merging the gradients, launching an optimization algorithm which involves executing modeling simulations and finally, updating the models with the merged gradients. On top of that, this is performed inside a minimization loop which is nested inside a frequency selection loop. This leads to complex interactions between nodes and therefore requires an extremely robust environment to be successful, in particular for large 3D inversions. Thus there is a challenge in terms of computational cost, but also of development cost, in order to obtain an FWI system which produces inverted models both reliably and efficiently.

Retrieving Elastic Parameters from Short Offset Geometry Acquisition

Full Waveform Inversion of elastic dataset is challenging due to the complexity introduced by free-surface effects or P-S waves conversions among others. In this context, large offset are prefered for inversion because they favour transmission modes which are more linearly related to P-wave velocity. In contrast, the problem addressed here is whether short offset inversion should be suitable for seismic imaging. To answer this question, we choose to invert an elastic dataset obtained from the SEG/Overthrust SEG/EAGE model, with simultaneous inversion of Lame parameters. We will show that by using short offset and with only minimal pre-processing of the data (e.g. low-pass filter), we are able to construct reliable gradients. This, in turns, leads to accurate inversion of both shear and compression velocity models.

Supporting Massive Parallelism in Seismic Processing

Traditionally, most of the literature in seismic processing area is devoted to proposals regarding the physics and numerical methods used for solving the underlying problem, such as: modellings, migrations, inversions, etc. There are, however, a lot of effort placed in the infrastructure needed to manage the processing carried out by such systems, as a lot of parallelism, fault tolerance, interaction, etc, is needed to actually produce a result. The aim of this work is to propose a framework, based on the Agile Modelling principle. Such framework, will let the developer create a modelling, migration or inversion process (among others), taking care of the topics that normally falls outside of the scope of seismic processing contributions.