Executive Summary

Overview

This proposal details a multi-year, multi-sponsor University-Industry research initiative on the application of advanced signal analysis and processing techniques to problems in oil and gas exploration and production. One of the striking features of seismic signals is their highly non-stationary character --- a property that is poorly dealt with by current analysis and processing tools. The central theme of the Rice Consortium on Computational Seismic Interpretation is the application of time-frequency representations and wavelet transforms to seismic and well-log signal analysis, interpretation, and processing. The initiative leverages 30+ years of leadership in signal processing research at Rice University towards two primary objectives: (1) systematic development of advanced time-frequency-based seismic attributes for enhanced feature extraction from multi-dimensional seismic data, and (2) application of wavelet-based signal processing tools to key problems in seismic and well-log data preprocessing. Technology transfer to the industrial sponsors will be achieved through software libraries (Seismic UNIX modules and Matlab code), interactive research meetings, focused collaborative work sessions, and technical reports, preprints and publications.

Motivation and Significance

Seismic imagery of the earth's subsurface is critical to all aspects of the oil and gas exploration and production process --- from the location of fields to their appraisal, development, and subsequent monitoring. In exploration, seismic images of the earth's subsurface are scrutinized by interpreters who search for patterns correlated to possible hydrocarbon reservoirs. Recently, 3D imaging technology has become a standard exploration tool, particularly in mature hydrocarbon provinces like the Gulf of Mexico and the North Sea. The seismic interpretation process has changed radically as a result. While previously interpreters dealt with large plots of 2D cross-sections of the earth, they now work on computers with 3D volumes comprising Gbytes of data. There exists a great need for advanced tools for sifting through these mountains of data for features indicative of hydrocarbons.

One of the most striking features of seismic and well-log signals is their highly non-stationary character. This non-stationarity confounds traditional data analysis and processing tools, such as time-invariant filtering and Fourier transform techniques. As a result, these tools offer less than optimal performance. Clearly, non-stationary signals dictate matched, non-stationary analysis and processing techniques.

The central theme of this research effort is the application of time-frequency representations and wavelet transforms to seismic data analysis, interpretation, and processing. Time-frequency and wavelet representations measure local (in time and/or space) changes in frequency and scale content of a signal. Representations like the wavelet transform, the short-time Fourier transform, and the Wigner distribution figure prominently in a host of different application areas, including data compression; image coding and analysis; communications; speech and acoustic signal processing; and modeling and understanding of the human hearing and vision systems.

Time-frequency and wavelet representations map signals to a time-frequency/scale domain that acts like a generalized (time-varying) Fourier domain. Thus, in addition to analyzing seismic data, time-frequency/scale representations have natural applications in data processing. The time-frequency signal representation in terms of transient wavelets rather than long duration plane waves will enable high-performance non-stationary signal and image processing for detection, classification, compression, denoising, deconvolution, etc.

Seismic attributes aid the quantitative interpretation of seismic data by extracting information on the nature of its non-stationarity. The increased quality and resolution of seismic data, allows the deployment of quantitative signal analysis and feature extraction algorithms. Robust and automated seismic attribute extraction is becoming increasingly important for information extraction. Many of the currently used attributes lack the robustness and geological/physical significance to live up to this task. We will develop new seismic attributes based on a set of sophisticated high resolution time-frequency analysis tools developed over the past number of years at Rice.

Objectives

Our multidisciplinary approach to computational seismic interpretation and processing is unique in that it builds a bridge between advanced digital signal processing techniques and their application in geophysics. Our primary objectives are twofold:

Advanced time-frequency representations for seismic data:

Using the time-frequency paradigm, we will derive novel attributes particularly suited for extracting features and highlighting anomalies in modern 3D and 4D seismic data sets. Measures to be investigated include volume attributes (dip, azimuth, continuity, correlation) and event-based attributes (extracted along or perpendicular to the prevailing dip).

We will develop improved variants of the classical complex trace attributes (such as instantaneous frequency, bandwidth, Q-factor, etc.) based on a suite of powerful new time-frequency representations developed at Rice. The high performance of these representations will naturally lead to attributes that are more accurate, indicative, robust, and rapid to compute than their classical counterparts.

Wavelet-based seismic data processing:

We do not propose to simply apply existing wavelet processing techniques to seismic and well-log data, but rather to develop fundamentally new seismic processing algorithms based on wavelets. We will develop wavelet systems that are tailor-made for seismic processing tasks, in the sense that they are designed to take the specific properties of seismic and well-log signals into account.

In the near term, we aim to leverage 30+ years of signal processing experience at Rice (including 8 years of time-frequency and wavelet analysis experience) into seismic interpretation and processing. In the long term, we will expand our effort to address the challenges associated with analysis and processing of 3D, 4D and 4C seismic data. In particular, we will concentrate on fast and robust algorithms for dealing with the huge data volumes involved.

A detailed description of our promising preliminary results, research objectives, and plans are included in Appendices A-C.

Impact

Expensive to acquire and often impossible to reacquire, seismic and well data is perhaps the most important asset of any oil company. Effective hydrocarbon exploration and production depends heavily on signal processing algorithms to extract the maximum possible amount of information from each data set. However, current tools for information extraction do not match the fundamental non-stationary character of seismic data, and information extraction performance suffers as a result. High resolution time-frequency representations provide a natural domain for analyzing and processing non-stationary seismic data. Our new seismic attributes have the potential to revolutionize seismic data interpretation, enabling human seismic interpreters to search effectively and efficiently through mountains of data for the critical non-stationarities that indicate potential hydrocarbons. Furthermore, non-stationary processing techniques will provide geophysicists with new opportunities for improving on traditional seismic signal preprocessing algorithms.

It could be said that up to the present wavelets and time-frequency methods have not delivered as promised and have, to a large degree, been a disappointment in geophysics applications. While a huge body of advanced time-frequency research has been developed in the signal processing community, the link with geophysics has not been made directly. Only an interdisciplinary team made up of both signal processing and geophysics researchers in collaboration with industry can realize the true potential of time-frequency methods in geophysics. Here at Rice we have assembled the core of such an interdisciplinary team; in conjunction with industry we can indeed deliver revolutionizing interpretation tools using advanced signal processing.