# window length selection for optimum slowness resolution of ... window length selection for optimum...

Post on 27-Mar-2020

2 views

Embed Size (px)

TRANSCRIPT

Window length selection for optimum slowness resolution of the local-slant-stack transform

Sergi Ventosa1, Carine Simon2, and Martin Schimmel3

ABSTRACT

One of the most critical decisions in the design of a local- slant-stack transform (LSST) is the selection of its aperture, or more precisely, the selection of the appropriate number of traces and their weighting coefficients for each slant stack. The challenge is to achieve a good compromise between the slowness and the spatial resolution. Conventionally, the window size is chosen in a more intuitive manner by visual inspection and some limited tests. We analyzed the LSST to establish rig- orous criteria for the window selection to achieve the optimum slowness and spatial resolution in the transformed domain for a given data set. For this purpose, we estimated the slowness re- solution in the LSST domain as a function of the spatial-window

bandwidth and of the spectral characteristics of the waves. For a wave with a given bandpass spectrum, the slowness resolution, the stopband attenuation, and the wavefront-tracking capability are determined by the spatial window. For narrowband signals, the spatial window must be larger than the stopband bandwidth divided by the desired slowness resolution and the central fre- quency of the band. For wideband signals, the window length is determined by the lowest frequency components. Much longer windows can only be used when the slowness and the amplitude variations of the wavefront trajectories are small. We validated our approach with a synthetic example and applied it to a wide- angle seismic profile to show the filter performance on real data in which the LSST-window length is determined in an auto- matic, data-adaptive manner.

INTRODUCTION

Several seismic-processing techniques, such as noise filtering and velocity analysis, rely on the fact that seismic measurements are laterally coherent. Interfering, nonstationary noise will hinder these algorithms. Two commonly used features along seismic pro- files are: the similarity along signal trajectories, which correspond to wavefronts, and the wave-propagation direction given by the ve- locity/slowness vector at the receiver. When trace density is high, we employ this high similarity/coherence to design filters depend- ing on apparent slowness — inplane component of the slowness vector — to increase the signal-to-noise (S/N) ratio. Mainstream filters for highly coherent signals are based on plane-

wave decomposition techniques. These techniques can be classified into several broad categories: (1) pie-slice f-k (frequency-wave- number) filters (see, e.g., Embree et al., 1963; Yilmaz and Doherty, 2001); (2) filters in the τ-p (intercept-time, slope, or slowness)

domain, also called linear Radon or slant-stack domains (Stoffa et al., 1981; Durrani and Bisset, 1984; Turner, 1990; Deans, 1993; Yilmaz and Doherty, 2001; Wilson and Guitton, 2007); (3) filters in the slowness-frequency (p-f) domain (Forbriger, 2003; Dev and McMechan, 2009); (4) filters based on eigenvalue decompositions (see, e.g., Vrabie et al., 2004, 2006); (5) filters in the frequency-offset (f-x) domain (see, e.g., Bekara and van der Baan, 2009); and (6), time-space prediction-error filtering (see, e.g., Guitton, 2005). Note the close relation among the first filters; the f-k transform and the slant-stack transform (SST) are related by the projection-slice theorem (e.g., Durrani and Bisset, 1984); and the p-f transform and the SST, by the Fourier transform. The slowness variations of the signals strongly constrain the level

of sparsity that plane-wave-decomposition techniques can achieve. To overcome this limitation, these techniques adopt the following two main approaches: (1) They are generalized to represent seismic signals in a more compact way when signal trajectories can be

Manuscript received by the Editor 7 October 2010; revised manuscript received 12 August 2011; published online 16 February 2012. 1Formerly IFP Energies Nouvelles, Rueil-Malmaison, France; presently Institut de physique du globe de Paris, Paris, France. E-mail: ventosa@ipgp.fr. 2Consejo Superior de Investigaciones Científicas (CSIC), Marine Technology Unit, Barcelona, Spain. E-mail: csimon@utm.csic.es. 3Consejo Superior de Investigaciones Científicas (CSIC), Institute of Earth Sciences Jaume Almera, Barcelona, Spain. E-mail: schimmel@ictja.csic.es.

© 2012 Society of Exploration Geophysicists. All rights reserved.

V31

GEOPHYSICS, VOL. 77, NO. 2 (MARCH-APRIL 2012); P. V31–V40, 9 FIGS., 1 TABLE. 10.1190/GEO2010-0326.1

Downloaded 18 Feb 2012 to 161.116.100.92. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/

modeled, or (2) they are adapted to be used in a local way. In surface seismics, the hyperbolic Radon transform is used in common-mid- point (CMP) gathers for multiple attenuation and noise removal (e.g., Yilmaz and Doherty, 2001). Trad et al. (2002, 2003) impose sparsity constraints to attain high-resolution Radon transforms for trace interpolation/regularization. Mann et al. (1999) (see also Jäger et al., 2001; Hertweck et al., 2007) propose the common-reflection- surface (CRS) stack, as a local-second-order-approximation alterna- tive, when CMP is not powerful enough or when its hyperbolic assumption is no longer valid. The LSST (Ottolini, 1983; Harlan et al., 1984; Bohlen et al., 2004; Shlivinski et al., 2005) is a local adaptation of the SST. This transform represents the data in the time-space-slowness domain, applying one low-aperture SST cen- tered on each trace. As a result, the spatial resolution increases at the cost of “slowness resolution,” with respect to the full-aperture SST. Note the term slowness resolution stands for “smallest interval mea- surable along the slowness axis in the LSST domain” and has little in common with the concept of slowness resolution of a velocity model in seismic migration or seismic tomography. Widespread LSST applications are local-adaptive filters, e.g., the

spatial-averaging filters on degree-of-polarization measures (Schimmel and Gallart, 2003, 2004), or the adaptive f-k filters of Duncan and Beresford (1994); and instantaneous slowness measures (Milkereit, 1987; van der Baan and Paul, 2000; Hu and Stoffa, 2009). Hence, the LSST is found in leading-edge algorithms, such as CRS, to estimate the first-order-approximation parameters or stereotomography (Billette and Lambaré, 1998; Billette et al., 2003; Lambaré, 2008) to measure local slope in the event picking. A key element of these applications is the measurement of the

local slope of the signal trajectory or instantaneous slowness. Fomel (2002, 2007a, b) and Schleicher et al. (2009) define slope-extraction techniques based on the derivative of the wavefield. In the LSST domain, this measurement is determined as a local maximum of the amplitude, envelope or coherence along the slowness axis. Our goal is to estimate or to remove coherent signals by locally

decomposing the seismic profiles in slowness using the LSST optimally. The main issues to solve in the design of a filter based on the LSST are the determination of the optimal aperture to locally distinguish the signals in the transformed domain and the choice of an appropriate synthesis operation of the filtered signals to the time- space domain. We use the good energy concentration of the LSST on seismic events to reduce the complexity of the synthesis opera- tion. Regarding the first issue, the LSST window determines the aperture. The space-slowness location of a seismic signal in the LSST domain cannot be simultaneously known with an arbitrary high precision; the choice of the window entails a tradeoff between both. The spatial and the slowness resolution compromise set a minimum limit on the LSSTwindow, whereas the coherency length (distance where signals remain similar) and the slope variations set a maximum. The coherency length generally decreases with increas- ing frequency due to subsurface structure complexities; in light of this dependency, Schimmel and Gallart (2007) adapt the window length with frequency. The main task, when an LSST-based filter is employed, is to find

a good tradeoff among key parameters, such as the slowness reso- lution, the interference and noise rejections, and the tracking cap- ability of slowness-varying events. Conventionally, the optimum window is intuitively designed by visual inspection of some limited tests. To make this task easier and objective, we establish rigorous

design criteria by analyzing the influence of the spatial-window bandwidth and the signal spectra on the slowness resolution of the LSST decomposition. As a result, we show that the optimum window can be designed beforehand, given the minimum global or local slowness resolution required to filter a seismic section. This work is organized as follows. We first introduce the LSST

and the adaptive filters based on this transform. We then focus on the slowness resolution estimation, stressing its main dependencies, and the synthesis of rigorous criteria on the LSST-window design. To conclude, we apply the new tools on synthetic and real data profiles and we discuss the main results.

METHODOLOGY

Local-slant-stack transform

The LSST of a seismic profile uðt; xÞ with a weighting function gðxÞ, also called spatial-window, is

vsðτ; xcÞ ¼ Z þ∞ −∞

gðx − xcÞuðpsðx − xcÞ þ τ; xÞdx; (1)

where τ denotes the delay, xc the offset in the transformed domain, and ps the slowness, with s its slowness index. Note that gðxÞ is smooth, has unit area, and is positive around zero; and that τ and xc are continuous whereas ps is discrete. In the discrete domain, the LSST of a s