FILTRAGEM SVD APLICADA À MELHORIA DO RASTREAMENTO DE HORIZONTES SÍSMICOS

No presente artigo apresentamos uma aplicação da filtragem SVD (Singular Value Decomposition) para o mapeamento automático de horizontes sísmicos. A filtragem SVD pode ser vista como um método de filtragem multicanal onde cada traço filtrado guarda certo grau de coerência com os traços imediatamente vizinhos. Esta filtragem preserva as relações de amplitude, fase e correlação espacial dos eventos sísmicos, ao tempo em que permite eliminar o ruído incoerente, normalmente associado aos últimos autovalores. A decomposição SVD é realizada sobre o subconjunto de traços vizinhos a cada traço da linha sísmica 2D ou de um volume 3D. O traço filtrado é obtido utilizando apenas alguns dos autovetores e autovalores associados. Ilustramos a aplicação do método sobre dados sísmicos terrestres. A melhoria da coerência dos eventos sísmicos permitiu maior robustez ao autotracking no mapeamento e interpretação automática dos horizontes sísmicos. A filtragem SVD é computacionalmente eficiente e tem o mérito de melhorar significativamente a coerência, a consistência e a continuidade dos eventos de reflexão facilitando muito o "trabalho", do tracker na busca de padrões no processo de autotracking.Keywords : mapeamento automático de horizontes; processamento sísmico; filtragem SVD; rastreamento de horizontes sísmicos.ABSTRACTWe present an application of a singular value decomposition (SVD) filtering approach to the automatic detection of seismic horizons. The SVD filtering approach may be seen as a multichannel filtering method where each filtered seismic trace retains the coherence of the neighbouring seismic traces. The SVD filtering preserves the amplitude and phase relations and reinforces the spacial correlation between seismic events, and at the same time it reduces the incoherent noise in data, which normally is associated to the last eigenvalues. The SVD decomposition is performed on each subset of traces around each trace of the original 2D or 3D seismic data. The filtered trace is obtained from the most important eigenvalues and eigenvectors. We illustrate the application of the new approach on 3D post-stack land seismic data. The improvement of the resultant coherence in the seismic reflected events allows for greater autotracking robustness during the automatic interpretation of the seismic horizons. The SVD filtering approach is computationally efficient and improves significantly the coherence, the consistency and the spacial continuity of the seismic events making easier the automatic detection of the commercial software in the search for patterns along the autotracking process.Keywords : automatic mapping of horizons; seismic processing; SVD filtering; tracking horizons seismic.

Semiautomated suppression of above-surface diffractions in GPR data

Diffractions from above-surface objects can be a major problem in the processing and interpretation of ground-penetrating radar (GPR) data. Whereas methods to reduce random and many other types of source-generated noise are available, the efficient suppression of above-surface diffractions (ASDs) continues to be challenging. We have developed a scheme for semiautomatically detecting and suppressing ASDs. Initially, an accurate representation of ASDs is obtained by (1) Stolt [Formula: see text] migrating the GPR data using the air velocity to focus ASDs, (2) multichannel filtering to minimize other signals, (3) setting an amplitude threshold that targets the high-amplitude ASDs and effectively eliminates other signals, and (4) Stolt [Formula: see text] demigrating the ASDs using the air velocity, and remigrating them using the ground velocity. By excluding the obliquity correction in the Stolt algorithms and avoiding intermediate amplitude scaling, we preserve the ASDs’ amplitude and phase information. The final stepinvolves subtracting this image of ASDs from a standard migrated version of the original data. This scheme, which includes some important extensions to a previously proposed method, makes it possible to semiautomatically process large volumes of GPR data characterized by numerous highly clustered and overlapping ASDs. The user has control over the tradeoff between ASD suppression and undesired removal of useful signal. It achieves nearly complete removal of ASDs in synthetic data and significant suppression in field data. Once ASDs have been suppressed, their influence can be reduced further by applying relatively gentle multichannel filters. It is not possible to remove line diffractions that resemble subhorizontal reflections or retrieve subsurface signals from data saturated by ASDs, such that some blank regions may be left after applying the suppression scheme. Nevertheless, subsequent processing and interpretation of the GPR data benefit significantly from the suppression of ASDs, which otherwise would clutter the final images.