Processing Near-Surface Seismic Reflection Data - A Primer

Processing Near-Surface Seismic-Reflection Data: A Primer Gregory S. Baker Course Notes Series Roger A. Young, Series Editor Society of Exploration Geophysicists Processing Near-Surface Seismic-Reflection Data: A Primer Gregory S. Baker Department of Geology State University of New York at Buffalo © 1999 by Gregory S. Baker All rights reserved. This book or portions thereof may not be reproduced in any form without permission in writing from the author. ii Contents Preface ...................................................................................................................iv 1 Introduction .............................................................................................................1 1.1 1.2 1.3 1.4 Organization......................................................................................................1 Near-Surface Reflection Seismology Background................................................2 Processing Objectives........................................................................................3 Demonstration Data from Kansas and England.....................................................5 1.4.1 Processing Goals ...................................................................................5 1.4.2 Geometry and Equipment .........................................................................7 2 Processing Steps and Associated Pitfalls ...............................................9 2.1 Processing Flowcharts........................................................................................9 2.1.1 General Shallow-Seismic Processing Flowchart ............................................9 2.1.2 Kansas Data Processing Flowchart ..........................................................10 2.1.3 England Data Processing Flowchart..........................................................11 2.2 Preprocessing.................................................................................................12 2.2.1 Data Input and Seismic Data Formats........................................................12 2.2.2 Computing Time Considerations ...............................................................14 2.2.3 Defining Geometry and Trace Headers ......................................................16 2.2.4 First Examination of the Data: Identifying Signal and Noise ............................17 2.3 Improving the Signal-to-Noise Ratio (S/N)...........................................................23 2.3.1 Killing Noisy Traces ...............................................................................24 2.3.2 Muting Coherent Noise ...........................................................................25 2.3.3 Elevation and Refraction Statics Corrections..............................................31 2.3.4 Filtering...............................................................................................36 2.3.5 Scaling ...............................................................................................42 2.4 Velocity...........................................................................................................45 2.4.1 Common-Midpoint (CMP) Sorting ..............................................................45 2.4.2 Velocity–A First Pass ............................................................................47 2.4.3 Velocity Analysis ..................................................................................48 2.4.4 Moveout Corrections .............................................................................57 2.4.5 Residual Statics Corrections ...................................................................60 2.5 Stacking and Verification...................................................................................63 2.5.1 Stacking .............................................................................................63 2.5.2 Analyzing Stack with Shot/CMP Gathers ...................................................64 2.5.3 Muting Coherent and Incoherent Noise ......................................................65 2.6 Displaying Seismic Data....................................................................................66 2.6.1 Muting Low-Fold Data.............................................................................66 2.6.2 Converting from Time to Depth .................................................................68 2.6.3 Displaying ...........................................................................................69 3 Discussion .............................................................................................................71 3.1 Applications of Migration...................................................................................71 3.2 Critical Processing Steps..................................................................................73 3.3 Avoiding Pitfalls................................................................................................74 Appendices ................................................................................................................75 A. References.................................................................................................................75 B. Data included on CD-ROM............................................................................................77 iii Preface The idea for this primer stemmed from the observation that more than ever before, people in academia and industry are using shallow-reflection seismology as a noninvasive tool for determining the physical properties and geometry of the upper 100 meters of the subsurface. Traditionally, workers using seismic methods for academic or industrial purposes have been familiar with the use of the techniques for hydrocarbon or crustal exploration. Advances in seismic methods have come largely from “deep” exploration seismology because the drive underlying that industry is profit maximization. Seismic techniques involved with imaging from hundreds to thousands of meters below the surface of the earth, therefore, have had substantial monetary support to obtain the best data possible. Major use of shallow-seismic techniques today, however, is in the environmental industry (for site characterization, contaminant detection, etc.) driven by minimizing cost. Therefore, if the availability of better data reveals a need for more expensive and extensive site cleanup, improving seismic data quality may be counter productive (to the industry) and may not obtain a high priority for funding and quality control. The goal of this primer is to provide a basic near-surface seismic-reflection processing guide for workers who have not had industry- or academic-supported training or guidance but wish to maintain the integrity of seismic imaging as a tool for near-surface exploration. This primer will focus on processing two small data sets using standard common-midpoint (CMP) processing and will include many significant processing pitfalls encountered in previous work. G.S.Baker May 1999 iv 1 IN T ROD U CT I ON 1.1 Organization This document can either be printed or followed as an Adobe Acrobat PDF file (included on the CD-ROM). As a PDF file, each item on the Contents page is linked to its corresponding location, and the user can return to Contents by clicking the Page Number at the bottomcenter of every page. Additionally, all references in this tutorial are linked to the appropriate location in Appendix A (the Reference list). The main portion of this tutorial is Section 2, Processing Steps and Associated Pitfalls, which focuses on individual processing procedures, each of which is broken into three parts: i. The Process • The basics of the process are discussed briefly with graphical examples when appropriate. ii. Applying the Process • The procedure is applied to the Kansas or England example data with emphasis on the most important parameters. iii. Pitfalls Associated with the Process • Potential processing pitfalls associated with the procedure are explained, using examples where applicable. The seismic data examples in this Primer were processed and displayed using Seismic Processing Workshop (Parallel Geoscience Corporation) on a Macintosh G3 computer. 1 1.2 Near-Surface Seismic-Reflection Data Processing The first published examples of seismic reflections detected shallower than 50 m are from Pakiser and others at the U.S. Geological Survey (Pakiser and Mabey, 1954; Pakiser et al., 1954; Pakiser and Warrick, 1956; Warrick and Winslow, 1960). Digital seismographs were not yet available, and the method was not strongly pursued until the late 1970s and early 1980s. A classic paper by Hunter et al. (1984) developed the optimum-window technique, first used for constant-offset surveying but now adopted in typical common-midpoint (CMP) processing. At that time, however, the high cost of seismographs, computers, and processing systems limited the seismic investigation of the shallow subsurface. The cost-per-channel in 1982 US dollars for a 12-bit seismograph (~66 dB of dynamic range) was about $5000. Today, the cost-per-channel in 1999 US dollars for a 24-bit seismograph (>100 dB of dynamic range) is about $1000. Thus, in the past 17 years the cost of seismographs has come down about an order of magnitude, factoring in the importance of dynamic range. A user can be fully equipped with a 48-channel, 24-bit seismograph, cables, and 40-Hz geophones for around US $70,000. And now, for less than US $5,000, one can purchase a high-performance microcomputer, laser printer, and seismic-data processing software. Shallow-reflection seismic surveying is now fairly commonplace in academia (for example, see Special Section - Shallow Seismic Reflection Papers, 1998, Geophysics; Near Surface Geophysics Special Issue, 1997, The Leading Edge) and in industry, as noted by the significant amount of use by the environmental industry. Processing shallow-seismic-reflection data is different from the seismic data processing done in hydrocarbon exploration (e.g., Miller, 1992; Black et al., 1994; Baker et al., 1998) and can be difficult for workers with little or no signal-processing background. Processing shallow-seismic reflection data is most precarious when nonreflective coherent events are generated on final stacked sections and can be misinterpreted as reflections. The goal of this tutorial is to attempt to reduce the number of misinterpretations by avoiding pitfalls and to advocate the adoption of a conservative approach to processing. Specifically, the processor should attempt to eliminate (or aviod generating) coherent noise events, even at the expense of a final image of poorer quality. 2 1.3 Processing Objectives The first and possibly most important step in processing shallow-reflection data is to determine objectives: Is the shallow reflection survey being performed to image a specific structure or stratigraphic relationship, or is it being performed to determine some overall generality about the physical properties of the subsurface? It is important to examine the objectives early and keep them in mind throughout processing. Typically, the clarity and resolution of the final product are of the highest importance. Processes that tend to generate artifacts while increasing coherency such as f-k filtering, trace mixing, or migration are appropriate for goal-oriented processing. However, these processes can generate significant coherent events that have no geologic basis in the geometry of the subsurface. Therefore, before performing any of the aforementioned processes, the seismicdata processor must be absolutely certain that the resulting coherent events can be correlated to reflections on the original shot or CMP gathers. A coherent event in a final stacked section that cannot be corroborated by supporting evidence from minimally processed shot or CMP gathers must be assumed to be artificial. The processor should adopt the precept that when reflection hyperbolas are not seen in at least part of the raw data (or with minimal frequency filtering or scaling), real corresponding reflections will not exist even “after sophisticated processing.” One must remember that sophisticated processing techniques are derived from the hydrocarbon exploration industry, whose goal is enhancing existing reflections, not revealing reflections where they are not initially visible. An example of how processing procedures can create coherent events with no substantiation in the subsurface is shown in Figures 1.1 and 1.2. The seismic reflection section in Fig. 1.1 was generated by applying a processing flow created for use on an unrelated data set. Figure 1.2 is a correctly processed section of the seismic reflection data used in Fig. 1.1. Note that the section in Fig. 1.1 has no apparent correlation with the section in Fig. 1.2, even though the same seismic data were used for each. 3 Trace Number 1 0.0 10 20 30 1 0.0 0.0 15 30 0.0 Angular unconformity? No unconformity Subhorizontal bedding 0.1 Time (s) Bedding dips to left? 0.1 0.1 0.1 Faults? No faulting 0.2 0.2 0.2 Figure 1.1 Seismic data processed using processing steps from a different seismic data flow. The section looks quite good and appears to be interpretable although it has no connection with the real subsurface. 0.2 Figure 1.2 Correctly processed seismic section of the data in Fig. 1.1. Reflections are discrete and subhorizontal. By examining Figure 1.1, which is made up of coherent events that have no basis in the subsurface, the importance of carefully identifying reflections throughout the entire process is clear. The processes used to generate Figure 1.1 were from a processing flow containing f-k filtering and trace mixing, both of which increased the coherency of the existing artifacts to the point of generating a bogus seismic section. 4 1.4 Demonstration Data from England and Kansas 1.4.1 Processing Goals Figure 1.4 is a final stacked section of the Kansas seismic data that will be used in this primer and which is included on the CD-ROM. Offset Along Profile (m) 5 17 30 42 54 Depth (m) 0 20 40 Figure 1.4 Final stacked depth section of the Kansas data. The goal of collecting the data was to determine the depth to, and geometry of, bedrock. The bedrock reflection at ~10 m was created by the interface between Holocene fluvial deposits and Paleozoic basement rock (inferred from nearby borehole information). The reflection is distinct, coherent, and shows the variability in the bedrock topography. This data set was collected to determine whether there was a steep-sided Pleistocene glacial meltwater channel carved into the bedrock, which was later buried by Holocene sediments. On the basis of these data, it was determined that a buried meltwater channel did not exist in this portion of the flood plain. 5 Offset Along Profile (m) 6 30 55 Two-Way Traveltime (s) 0.0 0.1 0.2 Figure 1.5 Final stacked section of the data from England used in this primer. Figure 1.5 is a final stacked section of the England seismic data that will also be used in this primer and is also included on the CD-ROM. The England data were collected in an alluvial valley (Thames River), and the goal was to image bedrock stratigraphy and examine possible structure and geometry. Establishing the depth to the reflections was not of primary interest. 6 1.4.2 Geometry and Equipment Kansas Demonstration Data The Kansas data were collected using a BISON 24096 96-channel, 24-bit seismograph with Mark Products L40A 100-Hz geophones. The source was a 30.06 rifle fired 25 cm downhole into a prepunched hole. The profile was on a slight incline (3.0-m elevation change over 100 m). The geometry of the Kansas experiment was fairly simple: 96 single geophones were deployed along the profile at 0.5-m spacing. The source began 1 m off the end of the line, and then was triggered at 1-m increments through the line until 53 m (Figure 1.6a). At that point, the first 48 channels were leapfrogged to the end of the line. The source continued through the rest of the line at 1-m increments until it was 1-m off the end (Figure 1.6b). Thus, the data consist of 74 shot gathers with 96 traces in each. The data were recorded for 500 ms with a sample interval of 0.25 ms. Additional information can be found in Appendix B. ∆ X Receiver locations 1-96 Source locations 1-53 ∆ X Figure 1.6a The geometry of the Kansas data prior to leapfrogging. ∆ Receiver locations 49-144 X Source locations 54-74 Figure 1.6b The geometry of the Kansas data after leapfrogging. 7 ∆ X England Demonstration Data The England demonstration data set was collected using a Geometrics 2401 24-channel, 15-bit seismograph with 100-Hz geophones. The source was a Buffalo gun (described in Pullan and MacAulay, 1987). The profile was in alluvial surface deposits on a nearly horizontal surface. The geometry of the England experiment was also simple: single geophones were deployed along the profile at 2.5-m spacings, and the source was 12.5 m off-end. This geometry was preserved by rolling the source and receivers along the profile for 20 shots at 2.5-m increments (Fig. 1.7). Thus, the data consist of 20 shot gathers with 24 traces in each. The data were recorded for 250 ms, with a sample interval of 0.5 ms. Additional information can be found in Appendix B. Source X ∆ Receiver Locations 1-24 ∆ 60m 12.5m 5 R o ll Figure 1.7 The geometry of the England data. Locations are 2.5 m apart. 8 2 PROCE S S I N G ST E P S AN D A S S OCI AT E D PI T F AL L S 2.1 Processing Flowcharts Processing flows are as widely variable as seismic data. In order to optimize any final stacked section, the flow must be tailored to the specific problems of the data, such as specific noise problems, a low signal-to-noise (S/N) ratio, etc. Below is a generalized processing flow followed by the specific flows used for the demonstration data. 2.1.1 General Shallow-Seismic Processing Flowchart Raw Data Input Raw Data Define Geometry First Examination of Data Preprocessed Data Kill Noisy Traces Mute Coherent Noise Statics Corrections Filtering & Scaling S/N Enhanced Data CMP Sort Velocity Analysis NMO Correction Residual Statics Stack Stacked Data Confirm Reflections AttenuateCoherent Noise Final Seismic Section Figure 2.1 General processing flow for shallow reflection data. 9 2.1.2 Kansas Data Processing Flowchart W A KE1001 SEG -2 Input RAW DATA Truncate Unused Data Land O bserver N otes TRUN CA TED Source Survey N otes Geom etry D efinition Survey N otes H EA D ERS K ill N oisy Traces Tim ing Shifts O ffset Sort O ffset Sorted Static Tim e Picks PREPRO C A m plitude Equalization EQ UA LIZED Elevation Static Shifts F-K Spectrum FKSpectrum 01 Elevation Info SHO Tw STA TICS FK M ute D efinition A pply F-K Filter SH O TFK CM P Sort CM PSO RTED M ute D efinition V elocity Function Sm oothed V elocity Function A pply M ute M UTED N M O Correction NM O A pply N M O -Stretch M ute N M O M UTED Frequency Filtering CM P Stack STA CK01 Rem ove Low -Fold STA CKFIN A L 10 Stretch M ute D efinition 2.1.3 England Data Processing Flowchart england2.sgy SEG -Y Input englandRA W Land O bserver N otes Survey N otes Geom etry D efinition Source Survey N otes england_01 V elocity Function K ill Traces england_preproc Tim e Picks Refraction Statics Elevation Statics A pply Static Shifts Refraction Statics Shifts englandSTA TICS M ute D efinition M ute Refractions Correct ResidualStatics Constant V elocity Stacks england_m uted englandCV Ss V elocity Function N M O Correction CM P Sort M ute D efinition englandNM O ed CM P Stack CM P Stack O ptim ization Statics england_stack england_teststack ResidualStatics Shifts M ute Early N oise englandstack01 Tim e V ariant Butterw orth A utom atic Gain Control englandstack02 Rem ove Low -Fold D ata EnglandFIN A Lstack 11 Q C Static Shifts englandN M O resid H orizon Guide 2.2 Preprocessing 2.2.1 Data Input and Seismic Data Formats i. The Process Most seismographs will record seismic data in SEG-Y (Barry et al., 1975) or SEG-2 (Pullan, 1990) seismic data formats. These are standard magnetic tape formats defined by the Society of Exploration Geophysicists (SEG). The difference between the two is the change in the size and format of the trace headers and in the storage of the trace-amplitude data. Seismic-data processing software packages, for philosophical or computational reasons, will normally convert from SEG standard formats to their own internal format for data processing and manipulation. Thus, the first step in a processing flow for shallow seismic data typically includes converting from SEG standard format to the specific format of the program. ii. Applying the Process This conversion process includes four important parameters. The first, data format, requires knowledge of the seismograph used to collect the data. The processor must know whether the seismograph records in (1) integer, (2) floating-point, or (3) compound format. Both example data sets were recorded in integer format. The three other important parameters, trace length, sample interval, and the number of channels, can be retrieved easily from good observer’s notes. For the Kansas data, the trace length was 0.5 s (500 ms), the sample interval was 0.25 ms, and 96 channels were recorded. The England data trace length was 0.25 s (250 ms), the sample interval was 0.5 ms, and 24 channels were recorded per record. iii. Pitfalls Associated with the Process Some seismic processing packages will perform a data format conversion by retrieving all of the important parameters from the trace headers without requiring user input. However, if the trace headers were not stored correctly in the field by the seismograph, or if the processing package does not automatically read the important parameters from the trace headers, the information must be provided by the data processor. Typically, a process in which the wrong data format is specified will not proceed to completion. However, specifying an incorrect trace length, sample interval, or number of channels will in most cases proceed to completion. When the processor specify a trace length 12 that is too long or too many channels, a stable processing package will insert zeroed data or blank traces, and the mistake will be obvious when first examining the seismic data. Conversely, the conversion process may proceed without error when the record length or number of channels specified is too small, and some information will be lost. A correct sampling interval is also critical. Even when an improper sampling interval does not cause the processing package to abort, all of the depth and velocity information gleaned from the seismic data during processing will be in error. The best way to confirm that the sampling interval is correct is to identify the air wave and make sure its velocity is ~335 m/s. The air wave is the energy that travels directly from the source to the receivers above ground; this energy is the audible sound generated by the source. Air wave is identified by three main criteria. First, air wave can be observed traveling across the seismogram without static time shifts from near-surface velocity changes. Thus, the air wave is usually the most coherent phase on a seismogram. Second, attenuation (or the inverse of attenuation, Q ) is typically different for air than it is for consolidated or unconsolidated material. Because Q is much higher for air, high frequencies are not attenuated as strongly as other seismic phases. Thus, air wave usually contains higher frequencies than other phases. The third way to identify air wave is to calculate the velocity of the apparent air-wave phase on one or more field files. It should be in the 330 to 340 m/s range. However, when the sampling rate is not entered correctly, the air-wave phase will yield a much different velocity. Thus, the best way to check the sampling interval is to use the first and second criteria described previously to identify air wave, and then check to see whether the velocity is as expected. 13 2.2.2 Computing Time Considerations i. The Process One positive attribute of processing shallow seismic reflection data is that the total size of the data set is typically small relative to exploration-scale data sets. Because of this, processors have the ability to perform several steps of processing at one sitting. Indeed, a processing flow that could take several days using a high-performance computer for exploration-scale data may be performed in hours or minutes on a desktop computer when dealing with shallow-seismic data. ii. Applying the Process For the Kansas data, 500 ms of data were collected. Collecting too much data in the field is preferable to collecting too little, but in this case a cursory look at the example data showed that the target bedrock reflection arrived at ~40-80 ms, and no visible reflections existed below that depth (and were auxiliary to the goal). Thus, we can truncate the total record length of the data from 500 ms to 200 ms. This will decrease computing time by almost 60%. As a result a process that normally would take 10 minutes to execute will now take only 4 minutes. Over the course of processing a more sizable data set, such time savings could prevent days of processing. Additionally, the processor might re-sample the seismic data to a larger time-sample interval to reduce size. An important limiting factor if re-sampling data is its frequency content. When seismic data are recorded using a sample interval of 0.125 ms, the maximum frequency that will be accurately recorded--the Nyquist frequency--is : Nyquist Frequency = 1 , 2 x (sample interval in seconds) or 4000 Hz. When resampling data, the processor would determine the maximum frequency content of that data. A typical convention is that there should be at least 8 points per wavelength for the highest frequency. Thus, if the maximum reflection frequency of the data were 200 Hz, the minimum sampling rate would be: Min. Rate = 1 8 x (Maximum Frequency in Hz) 14 , or 0.625 ms. Thus the data could (see Pitfalls, below) be re-sampled from 0.125 to 0.625 ms, cutting processing time by 80% (i.e., one new sample for every five old samples). The resampling process includes anti-aliasing filtering to prevent the high frequencies from aliasing to lower ones. iii. Pitfalls Associated with the Process Resampling is a simple process when the Nyquist frequency is kept in mind. Upon first examining the data, the highest frequency content of the reflections should be determined. Then, using the Nyquist concept, a minimum sampling frequency for resampling can be chosen. Source-to-Receiver Offset (m) 10 60 10 60 Time (s) 0.0 0.1 0.2 Before Resampling After Resampling Figure 2.4 An example field file from the England data before and after improper resampling. First examination of the England data showed that the maximum frequency content of the reflections was ~300 Hz. In Figure 2.4, an example field file was resampled from the original sampling interval of 0.25 ms to 2 ms. The resampled data (Figure 2.4, right panel) shows degraded reflections because the Nyquist frequency for the new sample interval is 250 Hz; therefore, much of the reflection information was lost. 15 2.2.3 Defining Geometry and Trace Headers i. The Process Defining the geometry and correctly filling the trace headers is an important, and often tedious, process. When the geometry and the trace headers are correctly defined and set, spatial and temporal processing steps become possible. When trace headers are not correctly set, the processing steps will not be properly executed. ii. Applying the Process Correctly filling the trace headers involves inserting into the proper locations information from the observers’ notes (OB notes) written in the field during the collection of the seismic data. Typically, a seismic data-processing software package will use a specific processing step to create the headers. iii. Pitfalls Associated with the Process Correct trace headers are important because when the headers are incorrect, the processing is likely to be incorrect. Spatial processes, such as f-k filtering and velocity determination, will be incorrect when the spatial relationships among the traces is not correct. Additionally, CMP sorting will be incorrect when the headers are not correct. 16 2.2.4 First Examination of the Data: Identifying Signal & Noise i. The Process After the trace headers have been correctly entered, the processor should always take the time for a detailed first examination of the data to identify specific problems, obvious reflections, and coherent noise. This sounds easy, but correctly identifying reflections (signal) from the onset of data processing is not always straightforward, and misidentification will lead to an incorrect seismic section. ii. Applying the Process Kansas Data Trace Number 1 10 20 30 40 50 60 70 80 90 0.0 Time (s) Reflection 0.1 Air Wave Reflection Ground Roll 0.2 Noisy Trace Noisy Trace Noisy Trace Figure 2.5 A first examination of the Kansas data, with some phases identified. A first look at a typical shot gather (unprocessed) from the Kansas data (Fig. 2.5) shows several distinct features. First, noisy traces are evident (see Section 2.3.1). The second prominent feature is the high-amplitude ground roll. Ground roll, which in vertical-component P17 wave seismic data is typically composed of Rayleigh waves, is identified by two main characteristics. First, ground roll has a slow phase velocity (steep slope). Wave-equation physics constrains the propagation velocity of Rayleigh waves as being slower than the direct S-wave, which in turn must be slower than direct P-waves. The propagation velocity of ground roll for a Poisson’s ratio of 1/4 is 54% of the P-wave velocity for a homogeneous, isotropic medium. The second characteristic of ground roll is that it is dispersive (i.e., shingled or ringy). Ground roll propagates along the surface, and the depth of material affected is directly dependent on the frequency of the ground roll. The high-frequency component of the ground roll interacts with the very-near-surface material, whereas lower-frequency ground roll interacts with deeper material as well as with shallow material. Therefore, ground roll will be dispersive when the near-surface velocity structure is variable with depth (typically increasing with depth) because different frequencies of ground roll will travel with varying velocities, depending on the particular average velocity being sampled. The third characteristic of ground roll is that it typically has a lower dominant frequency than near-surface refractions or reflections. Ground roll has a different frequency-dependent rate of attenuation than S-waves or P-waves. Therefore, for a given propagating distance, the highfrequency component of ground roll is attenuated much faster than the P-wave reflections or refrations and is recorded with a lower frequency content. The final two important features to identify are coherent noise and reflections. These will be discussed in the Pitfalls section. 18 England Data Trace Number 1 10 20 0.0 0.0 Reflection(?) Refraction Reflection(?) Time (sec) Reflection 0.1 0.1 0.2 0.2 Air Wave Noisy Trace Figure 2.6 A first examination of the England data. A first look at a typical unprocessed shot gather from the England data (Figure 2.6) shows features similar to the Kansas data (noisy traces and strong reflections), but it also shows a very strong refracted arrival and air wave. The air wave is a typical problem in shallow reflection data (see Section 2.3.2) and is identified because its velocity will always be 330 to 340 m/s (with variations due to elevation, air pressure, temperature, and wind). Because of the differences between the Kansas and England data, special considerations during processing will be necessary. The most critical step for both, however, is correctly identifying the reflections. 19 When first examining data, the initial step is to identify the main features (described for both example data sets above). The next step is to examine the data using various filters and gains to get a sense of features that might not be obvious on the raw data and to determine the frequency content of the signal (which will be useful when resampling; see Section 2.2.2). Following are several panels of the same field file from the Kansas data with various filters and gains applied, demonstrating the importance of this step. 1 20 Trace Number 40 60 80 AGC Gain Gain + Med.-Pass Filter 0.1 Gain + High-Pass Filter Gain + V. High-Pass Filter Gain + Low-Pass Filter Time (s) 0.0 Figure 2.7 Various filters and gains applied to a single field file from the Kansas data. The top-left panel is the same raw, unprocessed data shown in Fig. 2.6. The top-right panel is unfiltered data with an AGC gain applied. The remaining panels have the same AGC gain applied, but with different band-pass filters. Details of the newly observed features are shown in Fig. 2.8. Note the frequency content of the noisy traces. 20 Trace Number 1 10 20 30 40 50 60 70 80 90 Time (s) 0.0 0.1 0.2 Refraction Reflection Air Wave Direct Wave 1st Multiple Reflection Ground Roll Figure 2.8 Field file from the Kansas data with detailed identification of phases after filtering and gaining. The field file and processing are identical to Fig. 2.7, right-center panel. The source pulse in this data appears as a doublet (i.e., two positive peaks per phase), and the first peak is picked for interpretation. This is most evident on the direct wave, reflection, and refraction, and with reversed polarity in the first multiple reflection. 21 iii. Pitfalls Associated with the Process When identifying reflections, the processor must always remember that other forms of coherent noise such as aliased ground roll or air wave, diffraction energy, or random coherency may all look like reflection events. There are several checks to increase confidence in an apparent reflection event: 1) Reflections should be visible on several records without much processing. If the processor identifies a reflection-like event on only one shot gather and cannot find it on other shot gathers, it should be discounted. Often a noise event at the time of recording may generate an apparent reflection. It should be discounted, but not forgotten. Remember that a 48-trace shot gather will have one contributing trace in 48 CMP gathers. If the apparent reflection has a high enough amplitude (or is incorrectly enhanced by processing), it may stack and show up on 48 different traces on the final seismic section! 2) A true reflection should remain visible over a band of frequencies. Always use several frequency filters with slight variations in pass-band frequencies on a questionable reflection. If the apparent reflection is a product of aliasing, it will noticeably change its appearance for different frequency ranges. 3) Reflections should be hyperbolic, and this can be checked directly by fitting a hyperbolic curve through the event or picking three points on the event and calculating the fit. However, reflection events will not be truly hyperbolic when they are generated by an undulating surface, traveling through a strong, laterally-varying-velocity medium, or when severe elevation statics problems exist. Therefore, deviations from hyperbolic moveout can be observed. But remember, a non-hyperbolic reflection event from one of the aforementioned causes should also be visible on adjacent shot gathers. The most common error during the initial examination of the data is misinterpreting refractions as reflections. When this is done, the processor will typically process the data to enhance what is believed to be a reflection. Thus, correct segregation of reflections and refractions from the onset is perhaps the most critical process in all of shallow seismic data processing. 22 2.3 Improving the Signal-to-Noise Ratio (S/N) The goal of seismic processing is to enhance the signal-to-noise ratio (S/N) of the data. Three ways to improve S/N are: 1) Attenuating noise information in a domain in which signal and noise can be separated. Muting is a way of attenuating noise that has different traveltime and offset positions than reflections in the time-offset (t-x) domain. Frequency-wavenumber filtering is a way of attenuating noise that has a different spatial relationship (slope) than reflections. It is performed in the f-k domain. Frequency filtering is a way of attenuating noise that has a different frequency content than the reflections and is done in the amplitude-frequency (or frequency) domain. Each of these techniques assumes that S/N of the selection of data that is being muted is significantly lower than the remaining information. 2) Correcting for spatial or temporal shifts of the traces. Spatial shifts in the data are caused when the conditions in the subsurface violate the layered-earth assumption. These spatial shifts can be corrected using migration when sufficient velocity information about the region is known (see Section 3.1). Additionally, lateral velocity variations in the region above the water table (the weathered zone) create temporal shifts in the shot gathers such that a hyperbolic reflection event is distorted. Several correction techniques exist to compensate for this effect. However, seismic processing for shallow data typically is used to retain information from the weathered zone because it is within the region of interest. One type of temporal static that needs to be corrected in shallow processing is due to source and receiver elevation differences. Elevation statics are used to correct for temporal shifts caused by deviations from the datum plane of the source and receivers during the recording process. 3) Stacking. Theoretically, S/N increases as the square-root of the fold of the seismic data. This is based on the assumption that reflection information is embedded in random noise. Thus, during stacking, the signal will increase in amplitude by a factor equal to the fold due to constructive interference, and the random noise will sum to random noise with only slightly higher amplitude. The higher the fold of the seismic data, the higher the S/N. However, this assumption is typically violated by the addition of nonrandom (coherent) noise to the seismic data, in which case the S/N ratio will not increase as rapidly as the square-root of the fold and, in some cases in which the coherent noise is not properly removed, S/N will not increase at all or will decrease with increasing fold. Stacking is covered in Section 2.5. 23 2.3.1 Killing Noisy Traces i. The Process Simple but important, killing noisy traces should be one of the first processes applied to the data (see Pitfalls, below). The process of “killing” refers to setting to zero all of the amplitude values in a specified trace. ii. Applying the Process The noisy traces seen in Figures 2.5 and 2.6 could be selected and muted one at a time, but in most cases a noisy trace will be due to a bad connection or bad geophone at a particular receiver location that was not identified in the field. In this case, most processing packages allow for all of the traces from a particular receiver location to be zeroed quickly and easily. This was true for the England and Kansas data. iii. Pitfalls Associated with the Process Noisy traces must be killed for two reasons. First, even when a noisy trace appears to contain some reflection information, it still has a lower S/N than the rest of the data and will therefore only serve to decrease S/N of the final section. Removing any trace with a lower S/N is almost always better than assuming that important information will be lost if the trace is removed. The second and most important reason noisy traces should be killed is more subtle. Some noisy traces can contain data “spikes” in which a single sample has the maximum amplitude and the adjacent samples are much smaller. This creates two problems: First, the spike will appear to have infinite frequency and may cause frequency-related processes to behave badly. When frequency filtering is applied, the spike will be convolved with the filter operator and appear as a high-amplitude wavelet with the same frequency characteristics as the filter operator. Second, because the amplitude of the spike is anomalously high, it will not “stack out” under the assumption that it is random noise. Thus, if any process is applied that produces spatial effects on the data (trace mixing, f-k filtering, migration, etc.), the single spike will contaminate much more of the data; it may even appear as a continuous coherent event on a stacked section. 24 2.3.2 Muting Coherent Noise i. The Process A method for increasing S/N is to remove noise that has a different location than the signal in the t–x (or shot) domain. Specifically, properly muting refractions, air wave, and ground roll all increase S/N. For data in which an air-wave phase is dominant, a processor might consider spatially (f-k ) filtering the data to remove the linear air-wave event. However, air wave is typically a high-frequency (often 1 KHz or more), broad-band noise form, and is usually aliased (Figure 2.9, below); thus, f-k filtering the air wave is likely to degrade the data (by enhancing the aliased air wave) rather than improve the data. If the aliased air wave shown in Fig. 2.9 is not removed successfully by some other means, it will stack constructively during the stacking procedure and generate coherent noise on the final stacked section. The best alternative is to surgically mute the air wave (see Applying the Process). When muting in any domain (i.e, t-x, f-k, etc.) the edges of the muted region should be tapered. A taper is used so that sequential data does is not abruptly set to zero, but rather gradually is reduced. The size of the taper must be large enough to minimize processing artifacts that occur at the edge of the muted region but small enough not to obscure signal. True A ir W ave V elocity Refraction A pparent A liased A ir W ave V elocity Reflection Figure 2.9 Example seismic data showing aliasing of air wave. The true velocity of the air wave is fairly slow (steep slope), but the aliasing of the air wave yields events with an apparent velocity closer to that of the reflection (aliased slope). 25 The key to muting is removing the portion of data in which S/N is much lower than S/N of the rest of the data. For example, Figure 2.10 shows that the removal of information with the noise cone where S/N is low can significantly enhance S/N of the data, even if the mute region represents a significant portion of the data volume. Source-to-Receiver Offset (m) -228 88 -228 88 0.0 0.0 Time (sec) REFRACTION MUTE NOISE CONE MUTE 0.1 0.1 Figure 2.10 An example from Baker et al., 1998 of shallow seismic data in which all of the information within the noise cone is degraded by air wave of the same frequency content as the reflections and thus was muted. Additionally, refractions were muted. The result of muting such a large portion of the data can be surprising (Figure 2.11). Note that although some reflection information was included in the muted region, S/N of the muted region was too low to contribute any important information. Thus, following a conservative approach to avoid contaminating the final stacked section by coherent noise, the processor could attempt to mute all regions with low S/N, even if it includes a significant portion of the data. 26 Distance Along the Seismic Profile (m) 48 96 144 Processed Time (sec) 0.0 0.05 0.10 Processed plus noise-cone mute Time (sec) 0.0 0.05 0.10 Data contained in noise-cone mute Time (sec) 0.0 0.05 0.10 Figure 2.11 The results of the severe noise-cone mute shown in Figure 2.10. Note that some signal is contained in the muted portion (bottom panel of the stacks) but is not of sufficiently high S/N to be worth keeping (from Baker et al., 1998). 27 ii. Applying the Process England Data The dominant coherent noise in the England data is composed of air wave and refractions. The England data did not contain significant ground roll, and reflection information with good S/N was observed within the noise cone. Trace number Time (sec) 1 10 20 1 10 20 0.0 0.0 0.1 0.1 0.2 0.2 Before Mute After Mute Figure 2.12 A preprocessed shot gather from the England data before and after muting the air wave and refractions. The mute taper length is 8 ms. The two noisy traces (2 & 17) were also muted. The data are displayed with AGC (40-ms window) and a bandpass frequency filter (250-300 Hz with 12 dB/octave slopes). Note that a portion of the reflection at ~35 ms was muted at farther offsets. However, that portion of the reflection interferes with the first-arriving refraction and thus has a distorted shape that would degrade the stacking quality. 28 Kansas Data Muting coherent noise within the Kansas data was accomplished with only one top-mute per record. As previously mentioned, air wave propagates at a velocity of ~335 m/s. At the Kansas site, the near-surface unconsolidated material had a P-wave propagation velocity slower than the air wave. The reflection energy of interest, therefore, occurs below the air wave (examine Fig. 2.12 as a comparison). Thus, the coherent noise to be muted consisted of refractions, direct wave, and air wave, and is located above the reflection of interest. Figure 2.13 shows a preprocessed common-midpoint gather before muting, during the mute-picking process, and after muting. 1 10 Trace Number 20 30 40 Preprocessed Time (s) 0.0 0.1 Mute Pick Time (s) 0.0 0.1 Muted Record Time (s) 0.0 0.1 Figure 2.13 A single preprocessed CMP-sorted gather from the Kansas data, with mute picking shown and applied. The mute taper was 8 ms. 29 iii. Pitfalls Associated with the Process The pitfalls associated with muting in the t-x domain generally come from failure to mute coherent noise either properly or entirely. Applying the mute process itself is straightforward. Comparing the top and middle panels of Fig. 2.11 shows the effect of failing to remove coherent noise completely in an attempt to keep all signal. Following is an example of the England data, in which the refractions and air wave were not muted, demonstrating the creation of coherent artifacts. Source-to-Receiver Offset (m) 0 49 24 74 0.0 Time (s) Low-frequency stacked refractions 0.1 High-frequency stacked aliased air wave High-frequency stacked air wave 0.2 Figure 2.14 The England data processed without muting air wave or refractions. The stacked, aliased airwave is moveout related and observed on low-fold CMP gathers. Figure 2.14 shows the significant effects of not muting the coherent energy (compare with the muted result, Fig. 1.5). Refractions stack to form coherent events. One hint that refractions are being stacked is than frequency does not decrease with depth (i.e., low-frequency events are seen earlier than higher frequency reflections) as one would expect with normal frequencydependent attenuation. Also, note the presence of coherent and incoherent air-wave noise. 30 2.3.3 Elevation and Refraction Statics Corrections i. The Process Long-wavelength variation refers to variability with a wavelength typically greater than the spread length (i.e., the distance covered by one “spread” of geophones). Elevation and refraction statics corrections are time shifts applied to traces to adjust for long-wavelength changes in elevation, or in thickness and propagation velocity of the weathering zone at different sources and receivers. The weathering zone, however, is typically the region above the water table, and it is sometimes the region of interest in shallow-reflection seismology. Correcting for the weathering zone in shallow seismic data therefore often defeats the purpose of determining the structure of very-near-surface stratigraphy, as is the case in the Kansas data. The purpose of using the England data, however, was to examine subweathering stratigraphy; thus, refraction statics were used. Short-wavelength variations, whose wavelength variability is less than the spread length, may be corrected for by using residual statics corrections (see Section 2.4.5). ii. Applying the Process Elevation Statics. The topography at the site in England is minimal; thus, no elevation corrections were made. However, the Kansas data have an elevation change of 3.0 m from one end to the other. The standard method for correcting for elevation changes is to perform datum statics, whereby each source and receiver is shifted in time to a flat datum (Figure 2.15). Two possible positions for the datum are common: First, the datum could be located at the lowest portion of the profile. This works well because the velocity of the material above the datum can be measured directly by calculating the velocity of the directarriving energy. The second method locates the datum at the top of the highest location in the profiles. This method is used when reflection events might be observed between the highest location and the lowest location on the profile. However, calculating the optimum velocity to use for the “empty material” above the lower elevations to the upper datum is difficult and often requires a known, flat, horizontal reflection event in the subsurface to constrain the values. Alternatively, a sloping datum can be used to minimize the amount of shifting necessary along a profile having a relatively uniform sloping trend (Treadway et al., 1988). 31 Upper Datum Lower Datum Figure 2.15 Statics corrections are calculated to move the apparent positions of the sources and receivers to either the high datum (left), or the low datum (right). The arrows represent the apparent shift in depth (positive or negative) of the source/receiver positions from the ground surface to the datum. For the Kansas data, the datum elevation was chosen as a low datum because no reflections were evident above the datum level (at 275 m/s, about 10 ms). The velocity above the datum is calculated by measuring the velocity of the direct-arriving P-wave energy (Figure 2.16). The velocity of the direct arriving P-wave energy remained very consistent along the profile at ~275 m/s. If the velocity of the very-near-surface layer varied over the length of the line (confirmed by observing changes in the slope of the direct arrival), an algorithm to correct for the elevation statics using a variable velocity could be applied, or refraction statics could be used (see below). Direct Arrival 0 Source-to-Receiver Offset (m) 7 17 27 37 Time (s) 0.0 0.1 0.2 Figure 2.16 A portion of a shot gather from the Kansas data, with an example of a direct-arrival pick. The direct-arriving phase is measured to determine the velocity used for elevation statics. 32 Refraction Statics. When the target of a near-surface seismic reflection survey is below the weathering zone, long-wavelength variations of the thickness and velocity of the weathering zone create static shifts that may degrade the final stacked section. However, the static shifts can be partially corrected computationally by applying refraction statics corrections. In the simplest method, the assumption underlying this process is that any deviations of the firstarriving refractions from a straight line are caused by variations above the refracting interface, not by undulations in the interface itself. If the assumption is valid, source and receiver static shifts can be calculated from time picks of the first-arriving refraction. Shown in Figure 2.17 are the time picks of the first-arriving refraction that were used in the calculation of the static shifts during the refraction statics process. England Shot Gather19 Time (s) 0.0 0.1 Figure 2.17 A portion of a shot gather from the England data with the refraction time picks shown. Note the deviation from a straight line in the refraction at the far offsets. The far end of the England data profile was collected over a very-near-surface anomaly, which can be observed in the far offsets of the last five shot gathers (Figure 2.18). Note in Figure 2.18 that the static shifts caused by the very-near-surface variations affect the reflections as well, and without correction they would degrade the final stacked section. After refraction statics corrections are applied, such deviations are minimized. 33 England Shot Gathers 16-20 Raw England Data Time (s) 0.0 0.1 0.2 With Refraction Statics Time (s) 0.0 0.1 0.2 Figure 2.18 The last five shot gathers of the England data before (top) and after (bottom) refraction statics. Note the reduction of the time shifts in the refractions and reflections at the far offsets (e.g. circled regions). 34 iii. Pitfalls Associated with the Process As previously mentioned, when a low datum is used to calculate statics corrections, it is possible that reflection information will be shifted off the seismic section. The main problem with elevation corrections, however, is choosing the correct average velocity from the surface to the datum. If the velocity is incorrect, overcorrections or undercorrections will cause false structure in the final stacked section (Figure 2.19). Offset (m) 6 8 10.5 13 15.5 18 Offset (m) 20.5 23 25.5 6 8 10.5 13 15.5 18 20.5 23 25.5 0.1 Elevation Time (ms) 0.0 Over Corrected Under Corrected Figure 2.19 A portion of the England data showing the effects of under- and overcorrection of elevation statics due to incorrect velocity values. Typically, a plot of the elevation should be shown along with a stacked section to ensure that a viewer can locate apparent structure that may be due to incorrect elevation statics corrections. When the velocity values used for elevation statics corrections are incorrect, apparent structure will be proportional (overcorrected) or inversely proportional (undercorrected) to the elevation change across the profile. 35 2.3.4 Filtering i. The Process We have already discussed one type of filtering; i.e., muting in the t-x domain. Two other types of filters are used commonly: frequency-wavenumber (f-k) filtering and frequency filtering. Frequency-wavenumber filtering is a method in which linear coherent noise can be attenuated as long as the signal has a different spatial orientation (slope) than the noise. Filtering in the f-k domain can be useful in some instances (see Applying the Process) in which ground roll or other coherent noise cannot be attenuated without severely degrading signal quality; otherwise f-k filtering is not recommended (see Pitfalls). The other type of filtering, frequency filtering, is the mainstay of seismic processing. Frequency filtering is used to attenuate noise at frequencies other than the dominant frequency of the signal. When used with shallow-reflection data, this type of filtering is most often used to attenuate ground roll (Figure 2.20). Source-to-Receiver Offset (m) Time (sec) 20 0.0 30 40 50 60 70 80 90 20 30 40 50 60 70 80 90 0.1 Amplitude (dB down) 0 -40 0 400 800 0 Frequency (Hz) 400 800 Frequency (Hz) Figure 2.20 Unfiltered and frequency filtered shot gathers (top) with associated frequency spectrum (bottom) normalized to the maximum amplitude. The data on the right was band-pass filtered from 290 to 420 Hz with 12 dB/octave rolloff. 36 In Figure 2.20 ground roll (low-frequency, high-amplitude events at greater time than the refractions) has been attenuated by the frequency filter. Also, the filtered shot gather (right) now has observable air wave (the high-frequency, steeply-dipping event cutting across the reflections), which cannot be removed by frequency filtering. For shallow reflection data, air wave--which in exploration seismology is usually filtered with a high-cut filter--typically has a frequency content that overlaps signal and cannot be attenuated (also see Fig. 2.12). Another way in which the frequency filtering of shallow data is dissimilar to deeper seismic processing techniques is that inspection of the frequency spectrum is not an important tool for choosing filter parameters. Note that in Fig. 2.20 the dominant frequency of the reflections in the filtered data could not have been determined by looking at the frequency spectrum of the raw data. Trial-and-error is the best method to determine frequency filters for shallow reflection data, i.e., use the filter that best attenuates noise and brings out reflections. When using frequency filters on high-frequency shallow reflection data, consider that the maximum and minimum frequencies of signal have not been found until mostly noise is seen, so the processor should first use a wide range of filters to determine the entire frequency content of the signal. ii. Applying the Process No f-k filtering was applied to the England data. There was minimal unaliased coherent noise on the unstacked data and stacked data. Frequency-wavenumber filtering was not seen as a beneficial process because of the generation of artifacts. Processed Data 0 Kansas Data Amplitude (dB down) Raw Data -40 0 400 Frequency (Hz) 800 0 400 Frequency (Hz) 800 0 400 Frequency (Hz) 800 0 400 Frequency (Hz) 800 England Data Amplitude (dB down) 0 -40 Figure 2.21 Frequency spectra of the Kansas (top) and England (bottom) data. 37 Frequency filtering was applied to both data sets (Figure 2.21), although the results were not as dramatic as those seen in the example in Figure 2.20. The dominant frequency of signal in the Kansas data was ~100 Hz, and the dominant frequency of signal in the England data was ~210 Hz. Notice in Figure 2.21 that the result of processing, including frequency filtering, was to enhance the portion of the frequency spectrum containing signal for each data set. The Kansas data was amenable to f-k filtering for three reasons: First, S/N was fairly high for the reflection of interest, so good control of the f-k filtering parameters could be determined. Second, there was very little existing noise with the same spatial orientation (slope) as the reflection, so the noise would not be significantly enhanced by f-k filtering. And third, ground roll was coherent and not very dispersive and could therefore be f-k filtered with a fairly gentle taper. The separation of the phases in the f-k domain is shown in Fig. 2.22, the picking of the f-k mute region is shown in Fig. 2.23, and the filtered records are shown in Fig. 2.24. Normalized Wavenumber (1/λ) -1 .0 -0 .5 0.0 0.5 1.0 Frequency (Hz) 600 400 200 0 Air Wave Aliased Air Wave Direct Wave Reflection Ground Roll Lower-Order Ground Roll Mode Figure 2.22 An f-k spectrum (top) and an interpreted spectrum (bottom) for the Kansas data shot gather shown in Fig. 2.8. Coherent information sloping down from left to right plots with a negative wavenumber (left side of the spectrum). Notice that aliased air wave has an apparent opposite slope (i.e., a positive wavenumber). The segregation based on slope allows for f-k filtering to separate noise from signal. 38 Frequency (Hz) 400 300 200 100 0.0 -1 .0 -0 .5 0.5 0.0 1.0 Normalized Wavenumber (1/λ) Figure 2.23 A portion of the Kansas data f-k spectrum in Fig. 2.22 with the f-k-mute region shaded. The mute region should be symmetrical about the zero-wavenumber axis, regardless of the data, to avoid f-k artifacts. The mute taper was 20 Hz. Trace Number 1 15 30 45 60 75 90 Preprocessed Data Time (s) 0.0 0.1 f-k Filtered Data Time (s) 0.0 0.1 Figure 2.24 Kansas shot gather before (top) and after f-k filtering (bottom). The reflection (red dashed hyperbola) is greatly enhanced, and ground roll is attenuated with minimal artifacts. 39 iii. Pitfalls Associated with the Process Frequency-wavenumber filtering is a process that can generate linear coherent noise, which can be easily misinterpreted. Figure 2.25 shows the results of improperly applied f-k filtering on a stacked section of the England data. The critical parameters in f-k filtering are to pick a mute region that is symmetrical about the zero wavenumber axis and to have gentle tapering on the edges of the muted region. Offset (m) Time (ms) 0 6 8 10.5 13 15.5 18 20.5 23 25.5 100 200 Offset (m) 100 200 8 10.5 13 15.5 18 Offset (m) 20.5 23 25.5 0 Time (ms) Time (ms) 0 6 6 8 10.5 13 15.5 18 20.5 23 25.5 100 200 Figure 2.25 Partially processed England stack without f-k filtering (top), and the two poorly f-k filtered stacks (bottom left and right). Two distinctive characteristics suggest improper f-k filtering: a “wormy” appearance (bottom left), and broken coherent events all of which have the same positive or negative slope (~15°; circled bottom right). 40 The artifacts seen in the bottom panels of Figure 2.25 must be avoided. Notice in both cases that linear coherent noise has been generated: In the bottom left panel of Figure 2.25 coherent reflection events have been “smeared” into positions where they were not imaged, and in the bottom right panel dipping coherent events now appear in place of originally flat-lying reflections. Thus, it is critical that f-k filtering be used properly, as it is one of the processing steps most likely to generate artifacts that can lead to improper interpretation. Frequency filtering also has pitfalls. The most obvious pitfall would be filtering (removing) signal instead of noise and degrading S/N. Because the trial-and-error method is the best way to find the optimum filter for shallow reflection data, it is important to try all ranges of filters and not to assume a priori that the reflection information in your data will be in a particular frequency range. An additional pitfall is in designing band-pass filters in which the rolloff on either side of the pass band are too narrow (steep slopes). This will generate ringing–the Gibb’s effect–and will greatly distort the data (Figure 2.26). Slopes should typically not exceed 12 dB/octave. Source-to-Receiver Offset (m) 6 30 55 Source-to-Receiver Offset (m) 6 30 55 0.0 Time (s) 0.1 0.2 Figure 2.26 Partially processed England data (left) with gentle (12 dB/octave) tapers on the band-pass filter, and the same data with the same pass band, but with steep tapers (21 dB/octave). To avoid the Gibbs effect, filter slopes should not be steeper than 12 dB/octave. 41 2.3.5 Scaling i. The Process Scaling is the process whereby amplitudes of data are increased (or at least normalized) for display or processing purposes. Two factors affect the amplitudes of reflection data and associated noise: the amplitude of a spherical wavefront is inversely proportional to its distance from the source, and higher frequencies are attenuated faster than lower frequencies. Through the combination of these two effects, reflections (which typically have higher frequency and lower amplitude than ground roll and refractions) are attenuated with distance from the source. Gain corrections are used, therefore, to enhance the visible appearance of reflections, with a higher gain correction being needed for deeper reflections. Three main types of gain correction are constant gain, automatic gain control (AGC), and spherical divergence gain. ii. Applying the Process Both the England and Kansas data need gain corrections to make reflections visible. Because no amplitude analysis is to be performed, no spherical divergence correction is performed. Automatic-gain-control scaling is used in both cases (England data shown in Figure 2.27). Source-to-Receiver Offset (m) 1 10 20 1 10 20 0.0 Time (sec) 0.0 0.1 0.1 Figure 2.27 A portion of an England shot gather (after muting) without gain (left) and with AGC (right) using a 40-ms window length. Note the reflections when AGC is applied. 42 iii. Pitfalls Associated with the Process When any gain is applied, amplitude information is being altered. When true-amplitude analysis is to be performed, AGC must not be applied because AGC is not a reversible linear process, i.e., it is not possible to extract true-amplitude information from data that has had AGC applied. In all cases, too little or too much gain is a problem, but mainly only during the display of the data. The main pitfall associated with scaling comes from AGC. The critical (and often only) parameter specified in the AGC process is the length of the AGC window. This length is the time-band or window within which the amplitudes are normalized. One factor associated with window length is that, within the window, the highest amplitude information most strongly affects normalization. Thus, if the AGC window is too big (i.e., close to the record length), very little change in the data will be observed (most display features on processing packages already normalize each trace to its maximum amplitude). If the AGC window is too small (i.e., approaching the dominant period of the data), everything will be changed, and the data will be very noisy (Figure 2.28). Source-to-Receiver Offset (m) 1 10 20 1 0.0 10 20 0.0 Time (sec) Noise at Mute Taper 0.1 0.1 0.2 0.2 Late Arriving Reflections not Properly Gained Figure 2.28 The same England shot gather as in Figure 2.25, with a150-ms (left) and 5-ms (right) AGC window. An AGC window of 40 ms is shown in Fig 2.27. 43 Using too long an AGC window (Figure 2.28, left) is not a large problem, except that the processor will have poorly gained data. Using an AGC window that is too short causes signal to be overwhelmed by the enhanced noise. Additionally, very-small-amplitude processing artifacts at the tapered edges of mutes will be disproportionately increased as shown in Figure 2.28. Another problem with AGC scaling is that quiet zones, approximately equal to the length of the AGC window, can be generated by the AGC process both above and below high amplitude events. This is easily checked by varying the length of the AGC window and observing whether the size of the quiet zone changes. When the quiet zone varies with AGC window length, the AGC needs to be adjusted to minimize the effect, or a time-varying AGC window needs to be used. If the quiet zone remains, it should be addressed in the verbal description of the data. There are many instances in which an AGC quiet zone has been misinterpreted as a low-reflectivity zone, when in fact the subsurface geology has no relationship to the AGC quiet zone (Figure 2.29). 0 Time (ms) 80 160 240 320 20 ms AGC 40 ms AGC 60 ms AGC 80 ms AGC Figure 2.29 Four identical portions of stacked data. Each has had an AGC applied, with a different AGC window length. The high-amplitude event at the center of each section create AGC “shadows” or “quiet zones” in the incoherent noise above and below the event (bracketed by the purple lines). The quiet zones, generated solely by the AGC process, change in size depending on AGC window length. 44 2.4 Velocity 2.4.1 Common-Midpoint (CMP) Sorting i. The Process The concept of the common midpoint (CMP) is the basis for CMP processing. In the past, when seismographs with few channels and computational limitations were common, processing and manipulating shot gathers (in which the seismogram consists of all traces with a common source) to obtain an image of the subsurface was often successful because the strata being targeted were horizontal. With increasing use of reflection seismology and technology, however, spatial-processing algorithms applied to shot gathers began to break down when strata were not horizontal or when nonlayered objects were to be imaged. As a consequence, CMP processing was developed to take advantage of technology. The basis for success is that CMP gathers ideally image the same point in the subsurface, whereas shot 1s tS h 2n ot dS 3rd hot S etc hot . gathers represent multiple image points on the subsurface (Figure 2.30). *** ∆ ∆ ∆ ∆ ∆ *** ∆ ∆ ∆ ∆ ∆ ReSort Common Midpoint Shot Gathers CMP Gather Figure 2.30 Graphical depiction of how a simple re-sorting process can generate CMP gathers from shot gathers. Note that the paths included in the CMP gather represent only a subset of the information in the shot gathers. 45 ii. Applying the Process When the geometry-definition processing step has correctly generated the trace headers (the only possible pitfall), re-sorting traces from common-shot to CMP gathers is a simple reshuffling process. It is also possible to sort into common-receiver or common-offset gathers. 1 20 Trace Number 40 60 80 Time (s) 0.0 0.1 0.2 1 10 Trace Number 20 30 40 Time (s) 0.0 0.1 0.2 Figure 2.32 Three representative shot gathers (top) and CMP gathers (bottom) from the partially processed Kansas data. The shot gathers and CMP gathers have only several traces in common; however, similar phases can be identified on both. Except in cases of unusual geometry, CMP gathers will have fewer traces per record than shot gathers, typically by about a factor of two. Note that the fold of the Kansas data (fold is the number of traces per CMP gather) is variable along the profile. Variable fold is related to the way in which shot gathers map into CMP gathers. It occurs on most data collected as common-shotpoint data. 46 2.4.2 Velocity–A First Pass i. The Process At a point early in the processing flow, perhaps after preprocessing, when first-examining the data, it is important to get an idea of stacking velocities for the reflections in the data. These velocities will be used later to correct the moveout of the hyperbolic reflections (see Section 2.4.4). The first pass is important because it not only allows the processor to become more familiar with the data before more intensive processing begins, but it also gives the processor some reference points from which to start velocity analysis later. ii. Applying the Process A simple velocity determination can be applied if the processor has access to a processing package that allows the fitting of hyperbolas to the data. In Figure 2.33, hyperbolas are fit to shot gathers of the England data. Recall that fitting hyperbolas to shot gathers will only yield a correct stacking velocity when the subsurface is horizontally stratified and isotropic. If not, it will only be a starting point for later detailed velocity analysis. The other feature to identify during a first pass are multiples, which are arrivals of energy that have reflected several times in the subsurface. The simplest method, which can be used during the first pass and may lead to a modified processing strategy, is to identify whether any reflections have twice the zero-offset intercept time (t0) and the same moveout velocity. A possible candidate might be the first and second reflections in Figure 2.33, (0.36 x 2 = 0.72 ≈ 0.77), except that the velocities are ~20% different. When velocity is within ~5%, a multiple may be indicated and should be attenuated, either by predictive deconvolution (Sheriff and Geldart, 1995) or by choosing a stacking velocity that will not constructively stack the multiple. Trace number 1 10 20 1 10 20 Time (s) 0.0 t0 (sec) Velocity (m/s) .36 1524 .77 1941 1.21 2225 0.1 Figure 2.33 Preprocessed England shot gather without (left) and with (right) a preliminary velocity interpretation of the prominent reflections. This could be accomplished on CMP gathers equally well. 47 2.4.3 Velocity Analysis i. The Process If identifying reflections is the most important process, then velocity analysis is the second most important process. Reflections will not be correctly expressed on the final stacked section without correct velocity analysis. Velocity analysis is the process by which the stacking velocities of reflections are determined. The velocity-analysis procedure is essentially a forward-modeling procedure in which trial-and-error is the key to success. There are many methods of velocity analysis, such as using semblance velocity plots, sequentially picking hyperbolas on CMP gathers, using constant-velocity stacks (CVS), etc. Perhaps the easiest method to use with shallow-reflection data is CVS plots. Because shallow-seismic data are typically at least an order of magnitude smaller in data volume than exploration-scale seismic data, it becomes possible and fairly efficient to use constant-velocity stacks of the entire data set. Using the CVS method on the entire data set allows for the faster picking of velocities for maximum coherence on entire reflections rather than piecemeal velocity picking using other methods. ii. Applying the Process England Data Velocity analysis on the England data was conducted using CVS plots of the entire data set. Two things are important when generating CVS plots: First, the maximum and minimum boundaries of the NMO velocities should be sufficient to include the velocities picked on the first pass of velocity (Section 2.4.1) with generous overlap, typically +/- 30 %. Because shallow seismic surveys may include data from very-low-velocity near-surface material (<200 m/s) and higher-velocity bedrock (>1000 m/s) within the same section, it may be necessary to perform two CVS analyses to cover the range properly. For the England data, however, this was not necessary. The second important consideration when generating CVS plots is ensuring that the velocity step-size is small enough between sequential CVS plots (see Pitfalls). For some near-surface data, velocity increments as small as 5-10 m/s are necessary to get the best stack. Shown in Figure 2.34 are selected CVS panels from velocity analysis on the England data. In exploration-scale seismic data processing, the velocity picks can be chosen at the time of the reflection of interest. With shallow data, however, the velocity picks should be made to bracket the reflection with the NMO velocity of interest (see Fig. 2.35). The stacking velocities of sequential reflections on a single gather typically vary by more than 5% (e.g., Fig. 2.33); thus, because of automatic linear interpolation between velocity picks, a stacked wavelet may appear different than it did on the CVS plot when the NMO velocities for the top the bottom of the reflected wavelet are sufficiently disparate. 48 Velocity= 1400 Velocity= 1500 Velocity= 1600 Velocity= 1700 Velocity= 1800 Velocity= 1900 Velocity= 2000 Velocity= 2100 Velocity= 2200 Velocity= 2300 Velocity= 2400 Velocity= 2500 Velocity= 2600 Velocity= 2800 Velocity= 3000 Velocity= 3200 Velocity= 3400 Velocity= 3600 Time (s) 0.0 0.1 0.2 Figure 2.34 Several CVS panels from the England data. The top two rows are incremented by 100 m/s and the bottom row are in increments of 200 m/s from left to right. Each reflection becomes most coherent at different velocities, as expected. For the England data, CVS panels were generated and velocities picked in increments of 10 m/s (not shown). 49 Offset Along Profile (m) 6 30 55 Time (s) 0.0 0.1 0.2 Figure 2.35 Preliminary stacked section of the England data showing the timing of the velocity picks from CVS velocity analysis. Notice that the main reflections are bracketed rather than picked directly. The England data did not contain significant lateral velocity variations (Figure 2.36), and as a result smoothing was unnecessary, and picking velocities on the CVS plots at every fifth CMP gather was sufficient. It is important to output NMO-corrected CMP gathers after CVS velocity analysis before stacking to make certain that coherent stacked events are being generated by visible hyperbolas on CMP gathers. Velocity analysis using CVS plots is a time-saving method of analysis but requires this important quality-control step. 1200 m/s 3800 m/s Time (s) 0.040 0.080 0.120 0.160 5 10 15 20 25 30 CMP Number Figure 2.36 Stacking velocity field for the England data. 50 35 Kansas Data Velocity analysis of the Kansas Data was performed by fitting hyperbolas directly to the reflection of interest on CMP gathers. The Kansas data was analyzed at every fifth CMP because of lateral velocity variations. Several of the analyzed CMP gathers are shown in Figure 2.37. CMP loc. 32.0 CMP loc. 42.0 CMP loc. 52.0 CMP loc. 62.0 CMP loc. 72.0 CMP loc. 82.0 CMP loc. 92.0 CMP loc. 102.0 Time (s) 0.0 0.1 0.2 Figure 2.37 Analyzed CMP gathers from the Kansas data, with every 20th CMP gather shown. 51 Using this direct method of velocity analysis works best and is more efficient when only a few reflections are analyzed. With the Kansas data, the direct picking method is chosen because there is only one reflection. However, picking hyperbolas directly typically creates greater lateral variability in velocity picks (see Fig. 2.38) than if CVS or semblance methods are used. Therefore, lateral smoothing is necessary so that apparent discontinuities are not created by large NMO-velocity gradients. The Kansas velocity field was smoothed by two iterations of a three-point running average (Figure 2.39). Figure 2.40 shows the resultant smoothed velocity field. 330 m/s 400 m/s Time (s) 0.050 0.055 0.060 20 40 60 80 100 120 140 CMP Number Figure 2.38 A portion of the velocity field for the Kansas data prior to smoothing. The vertical black lines represent the locations of the analyzed CMP gathers. Note the shortwavelength variations due to picking variability. 400 N M O V elocity (m /s) Raw Picks 1st Iteration 380 2nd Iteration 360 340 320 300 20 40 60 80 100 120 CM P N um ber Figure 2.39 Graphical representation of the smoothing process performed on the Kansas data. The raw CMP hyperbola picks were smoothed using two iterations of a threepoint running average. 52 340 m/s 390 m/s Time (s) 0.050 0.055 0.060 20 40 60 80 100 120 140 CMP Number Figure 2.40 The velocity field for the Kansas data after smoothing. The vertical black lines represent the locations of the analyzed CMP gathers. Note that the short-wavelength variations due to picking variability have been smoothed. 53 iii. Pitfalls Associated with the Process There are two main pitfalls in velocity analysis. The first involves performing velocity analysis without corroborating coherent stacked events with unstacked reflections. This pitfall involves the improper identification of reflections as well as quality control shortcomings (see Section 2.2.4 and 2.5.2). The second pitfall in velocity analysis pertains specifically to shallow reflection data and relates to NMO velocities. As mentioned elsewhere, P-wave velocity within the upper 100 m of the subsurface can vary from as low as 130 m/s to as high as 5000 m/s. Normal-moveout corrections associated with large velocity gradients at the near surface may cause distortion effects (Miller and Xia, 1998). Additionally, near-surface data are very sensitive to small changes in the NMO velocity. To demonstrate this point, an example model (Fig. 2.41) is used to generate synthetic data (Fig. 2.42). Using the correct NMO and stacking velocity of 200 m/s, an accurate image of the model is produced (Fig. 2.43). However, when a velocity function with NMO velocity variations of only +/- 6 m/s is used (Fig. 2.44) to generate the stacked section, an apparent undulating reflection event results (Fig. 2.45). This example shows how apparent subsurface structure can be generated by using small variations in NMO velocity when none exist. Avoiding this pitfall in shallow-seismic velocity analysis, therefore, involves being sure that the lateral velocity gradients in very-low-velocity data are a result of careful identification of the true NMO velocity and not a function of CVS or semblance plots generated with velocity intervals too large. 0.5 m Depth (m) 0 * ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ ∆ P-wave Velocity 200 m/s 8 P-wave Velocity 500 m/s Figure 2.41 A velocity model used to generate synthetic data for demonstrating the effects of small changes in NMO velocity (see Figs. 2.42-2.45). 54 Source-to-Receiver Offset (m) 0 12.5 25.0 0.0 Time (ms) P-wave Reflection 1st Multiple Direct P-wave 1st Refraction Dispersive Groundroll Mode Conversions Model Artifact 0.1 0.2 Figure 2.42 Synthetic shot gather (left) and interpretation (right) generated from the velocity model in Fig. 2.41. Trace Number 1 50 100 150 200 250 300 Time (ms) 0.0 0.1 0.2 Figure 2.43 A synthetic stacked section using the data in Fig. 2.42 with a constant NMO velocity across the profile of 200 m/s. 55 194 m/s 206 m/s Time (ms) 0.0 0.1 0.2 1 50 100 150 200 250 300 Trace Number Figure 2.44 An NMO velocity function with variations of +/- 6 m/s across the profile. Trace Number 1 50 100 150 200 250 300 Time (ms) 0.0 0.1 0.2 Figure 2.45 A synthetic stacked section using the velocity function in Fig. 2.44 instead of the correct constant NMO velocity of 200 m/s. Note the significant variability that is generated using only small variations in NMO velocity. 56 2.4.4 Moveout Corrections i. The Process Moveout, or step-out, refers to traveltime differences between a recorded reflecting pulse as a function of shot-to-receiver distance and the true, vertical temporal location of the reflector at an offset distant of zero. Normal moveout (NMO) is the traveltime difference between the recorded traveltime for a reflection of a source-receiver pair with some separation and the traveltime from the same reflector if the source and receiver were at the same spatial position. Normal-moveout corrections are applied to data using the velocity field generated during velocity analysis. Another type of moveout, dip moveout (DMO), is the traveltime difference caused when recorded reflected events are generated by dipping reflectors not directly beneath the midpoint of a source and receiver pair. Dipping reflectors will generate both NMO (if the source and receiver are not coincident in space) and DMO. Both NMO and DMO corrections require accurate velocity information in order to be applied correctly. NMO and DMO effects must be corrected for, if they exist, to obtain a true spatial and temporal image of the subsurface. NMO Corrections Normal-moveout corrections are applied to CMP gathers to flatten hyperbolic reflection events for stacking. The basic assumption for determining NMO corrections is that user-measured stacking velocities are equal to the NMO velocities. This assumption is valid when structural dips or lateral velocity changes are small. The degree to which the assumption is valid determines the success of the CMP processing without more advanced procedures (see Section 3.1). The NMO velocity is the velocity at which hyperbolic reflection events will be correctly flattened. The NMO correction is applied to get the zero-offset traveltime to the reflection (t0): t02 = t x2 - x 2/V2rms, (1) where tx is the recorded traveltime to the reflection with the source and receiver at a horizontal offset x, and Vrms is the root-mean-square velocity to the interface. For the first reflecting interface, Equation (1) is simplified because Vrms is equal to the velocity of the surface layer, which can be measured from the direct-arriving energy on a CMP or walkaway seismogram (see Section 2.3.3). For later reflection events, Vrms can be approximated from stacking velocities on CMP gathers. Most seismic processing packages will apply the NMO correction algorithm given the velocity function determined from velocity analysis. 57 There are several more sophisticated processing steps that can correct nonhyperbolic moveout caused by either anisotropy or strong lateral velocity gradients (Yilmaz, 1987; Sheriff and Geldart, 1995). In most instances, however, anisotropy effects will not completely preclude a fairly good final stacked section, and the negative effects of strong lateral velocity gradients can be overcome by increased sampling of the subsurface (i.e., smaller station spacing; Baker et al., 1999). DMO Corrections Corrections for DMO are necessary when reflecting layers have significant dip. Dip moveout corrections are made in place of prestack migration. However, the DMO correction is negligible for small dip angles (< 10°) and is often not required in near-surface seismic data (see Section 3.1 for additional discussion). ii. Applying the Process When NMO corrections are applied, the amount of correction is related to depth (or time), NMO velocity, and source-to-receiver offset. The two cases in which NMO corrections are unstable are either when the velocity gradient is very large (Miller and Xia, 1998) or at sourceto-receiver offsets when NMO-stretch is an important issue (Miller, 1992). Reflection information can be distorted by differential NMO between the top and the bottom of the wavelet (Fig. 2.46). This NMO-stretch phenomenon is ubiquitous in near-surface CMP reflection imaging and can only be avoided by using other processing methods (see Section 3.1). However, an NMO-stretch mute can be applied during the NMO-correction process to mute the frequency-distorted reflection information, which will degrade the final stacked section. Source-to-Receiver Offset (m) 0 5 Source-to-Receiver Offset (m) 10 0 5 10 Time (s) 0.0 0.1 0.2 Before NMO Correction After NMO Correction Figure 2.46 A synthetic CMP gather without (left) and with (right) an NMO correction. Note that the near-offset wavelets after NMO retain a frequency similar to that of the uncorrected data, whereas the far-offset wavelets undergo NMO stretch and decrease in frequency by about a factor of two. 58 Both the Kansas data and England data have enough lateral coverage and the strata dips gently enough that DMO is unnecessary. However, both data have NMO-stretch problems at far offsets, and an NMO-stretch mute was applied. A good rule of thumb is to reject information that has stretch greater than 30% during NMO corrections. Miller (1998) suggested NMO-stretch mute as low as 10-15% when the high-frequency content of the reflections is more important than increased S/N. The England and Kansas data had 30% NMO-stretch mutes applied. This is in contrast to the common hydrocarbon-exploration practice of muting at 50% stretch. iii. Pitfalls Associated with the Process The main pitfall associated with NMO corrections is the NMO-stretch mute. When the mute is not severe enough, distorted low-frequency reflection information will be stacked into the final seismic section. Conversely, too severe an NMO-stretch mute can attenuate reflection information such that stacking will not increase S/N. Although a rule of thumb is suggested (30%), a trial-and-error analysis of the exact limit of the stretch mute should be performed for every data set. 59 2.4.5 Residual Statics Corrections i. The Process Elevation statics and refraction statics are used to account for long-wavelength variability where the wavelength of variation is greater than or equal to the receiver spread length (discussed in Section 2.3.3). Residual statics corrections, conversely, are used to adjust NMO-corrected data to account for near-surface variability with wavelengths less than the spread length. Typically, this variability is due to localized short-wavelength velocity variations, and because the variations may be less than a station spacing, it is impossible to reconstruct the actual geometry of the variations (i.e., they cannot be imaged using the existing experiment parameters). Therefore, the residual-statics-correction process is applied to NMO-corrected CMP gathers to adjust the traveltime deviations and improve the coherence of existing reflections. This process smooths the traveltime deviations using small time shifts within a user-specified limit. The basic assumption of the correction process is that near-surface variations can be adjusted for by shifting entire traces, not samples within traces. Additionally, residual statics corrections can be made while remaining surface consistent–a process that assumes time delays depend on the surface position of the sources and receivers and not on the path of energy in the subsurface. Both residual statics corrections and surface-consistent residual statics corrections, however, assume a near-vertical angle of incidence for the downgoing and upgoing energy at the sources and receivers. In exploration-scale seismic data, this condition is typically true because the velocity of the near-surface material is much slower than the velocity of the material of interest (Yilmaz, 1987). With near-surface seismic reflection data, this condition of near-vertical angles of incidence at the surface is not always true because (1) the lowvelocity near-surface is typically the volume being imaged, and (2) the spread-length to reflection-depth ratio is greater. Thus, the downgoing and upgoing raypaths at the sources and receivers are not vertical, and the process assumptions are invalid. Residual statics correction of both types therefore should be made with care, and often they will not be as robust as is observed in exploration-scale data. When residual statics corrections are applied, the final stacked section can usually be further improved by performing velocity analysis again after the residual statics are applied. And the entire procedure–velocity analysis, NMO correction, residual statics corrections–can be iterated until improvement in the data becomes negligible (see Fig. 2.1). 60 ii. Applying the Process After the first pass of velocity analysis and NMO corrections were made on the England data, the reflection events were observed to have residual short-wavelength variations Therefore, residual statics corrections were calculated and applied (Figure 2.47). Two iterations of velocity analysis, NMO correction, and residual statics were performed. Further iterations did not produce significant improvement in the data. However, applying the two iterations of residual statics corrections improved the final stacked section (Figure 2.48). England CMP Gathers Before Residual Statics Correction Time (s) 0.0 0.1 0.2 After Residual Statics Correction Time (s) 0.0 0.1 0.2 Figure 2.47 Representative NMO-corrected CMP gathers before (top) and after (bottom) application of residual statics corrections and 2nd-iteration velocity analysis. 61 England Stacked Data Time (s) 0.0 0.1 0.2 Before After Figure 2.48 Stacked England data before (left) and after (right) two iterations of velocity analysis and residual statics corrections. Note the significant improvement of the reflection at ~215 ms, which was barely visible prior to the process. iii. Pitfalls Associated with the Process Residual statics corrections should only be used on data with coherent reflections and a fairly high S/N ratio; otherwise, application of the process may generate coherent artifacts on a final stacked section. As with most other processes, the corrected CMP gathers and stacked section should be compared to CMP gathers prior to correction, and coherent events should be corroborated. 62 2.5 Stacking and Verification 2.5.1 Stacking i. The Process The process of stacking is straightforward: traces within NMO-corrected CMP gathers are summed vertically (stacked), and the subsequent traces are positioned at the CMP locations. Figure 2.49 shows a CMP gather from the England data, the resultant stacked trace, and its final position in the stacked section. CMP number CMP loc. 13.0 5.5 10 15 20 25 0.0 0.1 0.1 Stack Time (s) Time (s) 0.0 0.2 0.2 NMO-corrected gather CMP loc. 13.0 CMP loc. 13.0 Figure 2.49 An NMO-corrected CMP gather, stacked trace, and the position of the stacked trace in the final England stacked section. Note that without the adjacent traces, it would be very difficult to identify reflected energy on a single trace. 63 2.5.2 Analyzing Stack with Shot/CMP Gathers i. The Process This step is not only an intermediate step in the velocity-analysis process, but also a critical final step before the interpretation or publication of data. This is the stage in which original shot gathers, processed shot gathers, CMP gathers, NMO-corrected gathers, and the stacked section must be analyzed in an integrated way to ensure that the only coherent events on the stacked section are reflections. ii. Applying the Process Corroboration of reflections and coherent events on the stacked section of the England data was demonstrated previously (Figure 2.49). Figure 2.50 shows a processed shot gather, an NMO-corrected CMP gather, a stacked trace, and the Kansas stacked section. This corroboration was performed on both data sets, at multiple locations along the profiles. Shot Record 14 CMP number CMP loc. 27 12 0.0 36.5 61.5 86.5 111.5 0.1 0.1 0.2 Processed Shot Gather Processed NMO-corrected Stacked Trace CMP Gather Time (s) Time (s) 0.0 0.2 CMP loc. 27.0 Figure 2.50 A processed shot gather, an NMO-corrected CMP gather, and the Kansas stacked section. Similar displays were used all along the profile to corroborate coherent energy on the stacked section, with reflections observed on the shot gathers. 64 2.5.3 Muting Coherent and Incoherent Noise i. The Process After the major steps of the processing flow and the final analysis of coherent events have been completed, uncorroborated coherent and incoherent noise must be muted on the poststack seismic data. Post-stack muting is typically used to remove noise created during processing that has not been corroborated with the shot gathers. ii. Applying the Process A post-stack mute was applied to the final section of the England data (see Section 2.1.3). This mute removed minor distorted coherent events at the edge of the section and incoherent artifacts from previous filtering, scaling, and stacking (Figure 2.51). Offset Along Profile (m) 1 20 40 60 1 20 40 60 Time (s) 0.0 0.1 0.2 Before Post-Stack Mute After Post-Stack Mute Figure 2.51 England stacked section before (left) and after (right) muting noise. The muted portions (arrows) are most likely artifacts and noise from previous processing steps. iii. Pitfalls Associated with the Process The pitfall to avoid is failure to mute uncorroborated coherent or incoherent noise, either of which may be misinterpreted by those not familiar with the processing routine. As a result, the final seismic data also may be misinterpreted. 65 2.6 Displaying Seismic Data 2.6.1 Muting Low-Fold Data i. The Process Fold refers to the multiplicity of common-midpoint data. Theoretically, S/N will increase by (fold)1/2 if all stacked nonreflection energy is incoherent. Although the assumption that S/N increases with increasing fold is not always true, a strong relationship typically exists between fold and S/N. Shown in Figure 2.52 are the Kansas and England data with the associated fold diagram. Clearly, as fold decreases at the edges of the profile the S/N of the reflection events decreases. Muting the low-fold data at the edges of seismic sections will alleviate misinterpretation such as horizon truncations or pinch-outs by people not familiar with seismic processing. Additionally, fold will typically be variable in near-surface CMP data because time and equipment constraints often lead to field geometries which are untraditional by deeper exploration standards. For instance, the leapfrogging technique used on the Kansas data (see Section 1.4.2) created the uneven fold observed in Figure 2.52. If fold is not consistent along the main portion of a seismic profile, a fold diagram should be included in the final presentation to allow consideration of possible fold-related features on the section. Kansas Data England Data 0.0 Time (s) Time (s) 0.0 0.1 0.1 0.2 50 40 F O 30 L 20 D 10 0 15 0 F 10 O L D 5 25 50 75 Distance Along Profile (m) 0 0 25 50 75 Distance Along Profile (m) Figure 2.52 Stacked sections of the Kansas (left) and England (right) data with fold diagrams. Low-fold regions at the edges of section correlate with low S/N portions data. Note that the variable fold on the Kansas data does not correlate to any feature on the stacked section, and the geometry of the reflection event can be considered not related to fold. 66 ii. Applying the Process Determining the fold cutoff for the final stacked section depends on the fold variability, the S/N of the data, and the purpose of the data. The final sections of the Kansas and England data are show below with their associated fold diagrams (Figure 2.53). England Data 0.0 0.0 0.1 0.1 Time (s) Time (s) Kansas Data 0.2 40 F O 30 L 20 D 10 0 10 F 8 O 6 L D 4 2 25 50 75 Distance Along Profile (m) 0 25 50 Distance Along Profile (m) Figure 2.53 Stacked sections of the Kansas (left) and England (right) data with low-fold data removed. iii. Pitfalls Associated with the Process The two related pitfalls associated with removing low-fold data are removing too much or too little data. As mentioned previously, however, the amount to be removed is related to several factors, and no rule of thumb exists; it is possible for high-quality data to have no data removed at all. 67 2.6.2 Converting from Time to Depth i. The Process Often, one goal of collecting seismic reflection data is to determine the depth to a reflector or reflectors in the subsurface. When structural dips and lateral velocity gradients are small, seismic time sections can be converted to depth sections using the velocity function calculated during velocity analysis. The critical step is to convert stacking velocities to interval velocities using Dix’s equation (Dix, 1955). However, the degree of accuracy of the resulting interval velocities is directly affected by the accuracy of the stacking velocities and the amount of structural dip and lateral velocity variation (see Pitfalls). When interval velocities have been calculated, depth sections can be created by mapping the traveltime data sample-by-sample. At this point, external data such as refraction, borehole, or outcrop information can be used to better constrain the depth section, and interval velocities can be adjusted accordingly. ii. Applying the Process to the England and/or Vermont Data Depth conversion for the Kansas data was straightforward, as only one reflected interface was observed, and the interval velocity information could be calculated from the direct arrivals. Shown in Figure 2.54 is the depth-converted section of the Kansas data, without vertical exaggeration. The depth-to-bedrock was constrained by nearby outcrop information. Offset Along Profile (m) Depth (m) 5 0 14 24 34 44 54 64 10 20 Figure 2.54 Kansas data depth section with no vertical exaggeration, showing depth and geometry of the alluvium/bedrock contact. iii. Pitfalls Associated with the Process When using Dix’s equation to create a depth section by converting stacking velocities to interval velocities, errors will result when the aforementioned assumptions are invalid. If lateral velocity gradients or steep dips are expected, depth migration is the appropriate recourse; however, a very accurate velocity model is required (see Section 3.1). 68 2.6.3 Displaying Seismic Data i. The Process The methods used to display near-surface seismic data are the same as in other explorationscale processing, and the same critical points apply: sections should be labeled as time or depth sections; depth sections should be displayed without vertical exaggeration, if possible; and the type of display should reflect the purpose of the section. The five main types of display are wiggle trace, variable area, variable density, variable area wiggle trace, and variable density wiggle trace. Additionally, color or grayscale may be used for each type. Up to this point, all figures in the primer have been variable area wiggle trace displays. ii. Applying the Process Shown in Figure 2.55 and 2.56 are several methods of display and aspect ratios for the England and Kansas data. Coherency and detail are opposing factors in most cases. (c) (d) (a) (b) (f) (e) Figure 2.55 Several types of displays and aspect ratios for the England data: (a) variable area wiggle trace; (b) variable density; (c) variable area wiggle trace; (d) color variable density; (e) variable area; and (f) wiggle trace. There are no processing differences among the displays. 69 (a) (b) (d) (c) (e) Figure 2.56 Several types of displays and aspect ratios for the Kansas data: (a) variable area wiggle trace; (b) variable density; (c) wiggle trace; (d) variable area wiggle trace; and (e) color variable density. There are no processing differences among the displays. 70 3 D I S CU S S I ON 3.1 Applications of Migration Migration is useful when lateral velocity variations or reflection dip angles are severe, or when there is a large amount of diffracted energy. The migration process was developed for use on hydrocarbon-exploration data in which the number of channels per record, the number of traces per stack, and the S/N are high. Rarely are these conditions satisfied for near-surface data. The size of the data set is important because migration is an inversion process, and it generates numerical artifacts at the edges of data. The S/N ratio is likewise important for the migration process because noise can generate substantial artifacts upon inversion. Black and others (1994) considered the use of migration on near-surface data and concluded that because reflection depths are typically small compared with spread length and propagation velocities are low, the usefulness of migration is limited. For completeness, however, a brief overview of migration and references for further information follow. When the subsurface is composed of flat-lying homogeneous layers, stacked CMP gathers that have been NMO-corrected to represent zero-offset data will present a true temporal image of the subsurface. If, however, the subsurface consists of substantial lateral velocity variations or steeply dipping interfaces, information recorded from a surface receiver will not represent the structure directly beneath the receiver. Additionally, diffractions may occur on recorded data. Diffractions appear as hyperbolic events resulting when wave energy impinges on sharp objects or penetrates areas restricted by geometrical optics, and may obscure reflections and disrupt reflection continuity. Migration is an inversion process that can be applied to CMP data, either before or after stacking, that ideally will reassign reflection data to its true (zero-offset) position. Thus, the apparent angle of dipping reflections seen on unmigrated stacked data will be adjusted to the true dip angle of the reflector in the subsurface. Additionally, data can be converted correctly from time to depth, even when there are substantial lateral velocity variations. A summary of the geometric configuration involved in migration is given in Sheriff and Geldart (1995). Migration can also increase S/N of the data by removing or reducing the effect of diffractions, if they exist, by collapsing their energy to a single point. Migration works best when recorded events do not interfere (either constructively or destructively) with one another. Short-period noise may result in a smeared effect (migration smiles) on migrated data and usually occurs near the edges of the data, where S/N is reduced. 71 Current migration methods fall into two broad categories: diffraction-stack migration and transformation migration. Each migration method is based on slightly different assumptions concerning the data to be migrated. All migration algorithms, however, require accurate velocity information, which can be obtained by the methods previously described (see Section 2.4.2). Diffraction-stack migration techniques are based on work by Hagedoorn (1954). This method treats each reflector as a sequence of closely spaced diffracting points (Sheriff and Geldart, 1995). The most commonly used form of diffraction-stack migration is Kirchoff migration, which uses an integral solution to the wave equation (Schneider, 1978). This method adjusts amplitude values based on obliquity and divergence before transposing them from their apparent position. Similar migration methods include wave-front smearing and commontangent (Sheriff, 1978). Transformation migration techniques include frequency-wavenumber (Robinson, 1983), Fourier-transform or Stolt migration (Stolt, 1978), phase-shifting or Gazdag migration (Gazdag, 1978; Gazdag and Sguazzero, 1984a; Gazdag and Sguazzero, 1984b), and Radontransformation or slant-stack migration (Hubral, 1980). All of these techniques transform the data from the offset-time (x-t) domain to another domain prior to migration. When migration is successful, two factors are important: 1) the size of the data set must be sufficiently large to minimize the influence of numerical artifacts that occur at the edges of the data, and 2) time and resources must be available to enable the rigorous process of obtaining accurate velocity profiles and continuing through the iterative quality-control process that ensures the correctness of the results. 72 3.2 Critical Processing Steps All of the processing steps mentioned in this tutorial have the capacity either to generate the appearance of coherent noise or to improve the data quality. Thus, all are important. However, there are several steps that are of particular importance: 1) Initial Muting of Coherent Noise: This step requires that the processor have a good understanding of what constitutes signal and what is noise. The more noise that can be removed early on, the better the results of subsequent processing steps. 2) Velocity Analysis: The velocity analysis, and subsequent velocity field, are critical to producing a quality final stacked section. Failure to represent velocity gradients and signal coherency can result in a velocity field that can attenuate existing reflection events or create coherent noise when applied. Additionally, the accuracy of the velocity function is directly related to the quality of a depth-converted section. 3) Confirmation: This is the most important step in shallow-seismic data processing. It is relatively easy to generate coherent events on a stacked section that have no relation to the subsurface. Misinterpretation of poorly processed shallow-seismic data not only creates immediate problems that are directly related to the seismic data, but also adds to the already strong “black-box” view of seismic imaging. 73 3.3 Avoiding Pitfalls Various processing pitfalls have been detailed in this primer. However, each processing flow has unique pitfalls that can result in lowering the S/N ratio of the data or creating coherent noise that may be misinterpreted. As a result, a working knowledge of the particular pitfalls associated with the individual processing steps is critical to becoming a good overall shallowreflection seismic-data processor. Steeples and Miller (1998) have written an excellent paper specifically detailing pitfalls associated with near-surface reflection seismology. 74 A PPEN D I CES A. References Baker, G.S., Steeples, D.W., and Drake, M., 1998, Muting the noise cone in near-surface reflection data: An example from southeastern Kansas: Geophysics, vol 63, 1332-1338. Baker, G.S., Schmeissner, C., Steeples, D.W., and Plumb, R.G., 1999a, Seismic reflections from depths of less than two meters: Geophys. Res. Lett., 26 , no 2 , 279-282. Barry, K.M., Cavers, D.A., and Kneale, C.W., 1975, Recommended standards for digital tape formats: Geophysics, vol 40, 344-352. Black, R., Steeples, D.W., and Miller, R.D., 1994, Migration of shallow reflection data: Geophysics, vol 59, 402-410. Dix, C. H., 1955. Seismic velocities from surface measurements: Geophysics vol 20, 68-86. Gazdag, J., 1978, Wave equation migration with the phase-shift method: Geophysics, vol 43, 1342-1351. Gazdag, J., and P. Squazzero, 1984a, Migration of seismic data: Proceedings of the IEEE 72, No. 10, 1302-1315. Gazdag, J., and P. Squazzero, 1984b, Migration of seismic data by phase-shift plus interpolation: Geophysics, vol 49, 124-131. Hagedoorn, J. G., 1954, A process of seismic reflection interpretation: Geophysical Prospecting, vol 2, 85-127. Hubral, P., 1980, Slant stack migration: Geophysical Prospecting, vol 25, 728-745. Hunter, J.A., Pullan S.E., Burns, R.A., Gagne, R.M., and Good, R.S., 1984, Shallow seismic reflection mapping of the overburden-bedrock interface with the engineering seismograph-Some simple techniques: Geophysics, vol 49, 1381-1385. Miller, R.D., 1992, Normal moveout stretch mute on shallow-reflection data: Geophysics, vol 57, 1502-1507. Miller, R.D., and Xia, J., 1998, Large near-surface velocity gradients on shallow seismic reflection data: Geophysics, vol 63, 1348-1356. Pakiser, L.C., and Mabey, D.R., 1954, Mapping shallow horizons with reflection seismograph: Science, vol 119, 740. 75 Pakiser, L.C., Mabey, D.R., and Warrick, R.E., 1954, Mapping shallow horizons with reflection seismograph: AAPG Bull., vol 38, 2382-2394. Pakiser, L.C., and Warrick, R.E., 1956, A preliminary evaluation of the shallow reflection seismograph: Geophysics, vol 21, 388-405. Pullan, S.E., 1990, Recommended standard for seismic (/radar) data files in the personal computer environment: Geophysics, vol 55, 1260-1271. Pullan, S. E., and MacAulay, H.A., 1987, An in-hole shotgun source for engineering seismic surveys: Geophysics, 52, 985-996. Robinson, E.A., 1983, Migration of geophysical data: International Human Resources Development Corp., Boston, MA. Schneider, W. A., 1978, Integral formulation for migration in two and three dimensions: Geophysics, vol 43, 961-980. Sheriff, R.E., 1978, A first course in geophysical exploration geophysics: International Human Resources Development Corp., Boston, MA. Sheriff, R.E. and Geldart, L.P., 1995, Exploration Seismology, 2nd Edition: Cambridge University Press, New York, 592 p. Steeples, D.W. and Miller, R.D., 1998, Avoiding pitfalls in shallow seismic reflection surveys: Geophysics, vol 63, 1213-1224. Stolt, R., H., 1978, Migration by Fourier transform: Geophysics, vol 43, 23-48. Treadway, J. A., Steeples, D. W., and Miller, R .D., 1988, Shallow seismic study of a fault scarp near Borah Peak, Idaho: J. Geophys. Res., vol 93, 6325-6337. Warrick, R.E., and Winslow, J.D., 1960, Application of seismic methods to a groundwater problem in northeastern Ohio: Geophysics, vol 25, 505-519. 76 B. Data included on CD-ROM Five main components are included on the CD-ROM: 1) A file [READ.ME] that is an ASCII text file containing information about the contents of the CD-ROM. 2) A folder [Acrobat] that contains Adobe Acrobat 3.0 reader (freeware) information and installers for Macintosh, IBM compatible (Windows), and UNIX platforms. Acrobat can be used to interactively navigate through the digital version of this dissertation. 3) A file [Primer.pdf] that contains this document in Adobe Acrobat 3.0 PDF format. Each item on the Contents page is linked to the appropriate location within the document. Also, the page number at the bottom-center of each page is linked back to the Contents page. 4) A folder [ENGLAND] that contains a READ.ME file with details, raw SEGY data, raw SEGY data with correct header information, and a final stacked section (SEGY). 5) A folder [KANSAS] that contains a READ.ME file with details, raw SEG2 data, raw SEGY data with correct header information, and a final stacked section (SEGY). 77