Professional Documents
Culture Documents
TRUE DEPTH CONVERSION: More Than Preety Picture
TRUE DEPTH CONVERSION: More Than Preety Picture
Continued on Page 12
November, 2001
CSEG Recorder
11
ARTICLE
Continued from Page 11
Contd
Figure 2. a) Syncline feature in depth. b) Resultant bow tie reflection event in time (before migration). c) Seismic section after migration. d) Before migration. Migration is a seismic processing step to reposition reflections under their correct surface location (Figure 2).
Table 1
Continued on Page 13
12 CSEG Recorder November, 2001
ARTICLE
TRUE DEPTH CONVERSION: MORE THAN A PRETTY PICTURE
Continued from Page 12
Contd
Migration collapses diffractions. Migration puts the reflected energy back where it came from. These simple geometric examples listed above show that the need for migration arises when reflectors are dipping. The need for migration also arises when the subsurface velocities vary laterally, as variations in velocity will also cause reflections to be recorded at surface positions different from the subsurface positions. Time migration is strictly valid only for vertically varying velocity; it does not account for ray bending at interfaces. Depth migration accounts for ray bending at interfaces but requires an accurate velocity model. Depth migration is typically called for when there is significant lateral variation of velocities. Imaging addresses the proper lateral positioning of reflectors, but does not result in a true depth data set, even if depth migration is used (Al-Chalabi, 1994; Schultz, 1999). Depth migration depths often mistie known well depths; errors of over 100 metres are still common after depth migration (Haskey et al., 1998). The depth in depth migration is not true depth. Why? Because provelocities, those that do the best job of NMO and migration, are not the same as true vertical propagation velocities. Seismic energy, after all, does not travel vertically. There is a strong horizontal element to the travel path of energy that we record in any seismic surface data (Reilly, 1993; Schultz, 1999). Even if you do a zero-offset survey, and you send the source signal down vertically, the raypaths refract in accordance with Snells law whenever velocity variations are encountered. Because of Snells law and ray-bending, the signal that departed vertically will be unlikely to travel vertically. It is compelled to travel along at directions that are bent away from vertical. Nonetheless, provelocities are the right values to use for imaging. What makes them so fit for their purpose, though, makes them unfit for the purpose of true depth conversion, because they are designed to correct a different problem. You neither want to use vertical propagation velocities to do depth migration, nor use provelocities to do depth conversion (Table 1).
Is this unsettling? Intuitively, geophysicists feel that there must be an actual velocity at which the seismic wavefront travels through the ground. Over the years, though, velocity terminology has suffered casual use and often misuse.
Continued on Page 14
CGG Canada Services Ltd. Suite 700, 404-6th Avenue S.W. Calgary, Alberta T2P 0R9 (403) 266-1011 rvesely@ca.cgg.com
www.cgg.com
November, 2001
CSEG Recorder
13
ARTICLE
Continued from Page 13
Contd
Figure 3: A good approach to depth conversion. This will accomplish lateral positioning and vertical positioning to achieve actual depths.
14
CSEG Recorder
November, 2001
ARTICLE
TRUE DEPTH CONVERSION: MORE THAN A PRETTY PICTURE
Continued from Page 14
Contd
Figure 4: The paradox of velocity modeling: (a) when one minimizes the error at the tie point (direct conversion) the model can become unrepresentative of the true function, but when one honors the true function the error at the tie point increases (b).
November, 2001
CSEG Recorder
15
ARTICLE
Continued from Page 15
Contd
Figure 5
Adding more detail, we can move on to using interval velocities. Here, we assign a constant velocity to each layer within a given well (Figure 5b). Using average or interval velocities allows spatial variation of velocity between well locations. We can accomplish this by cross-plotting interval velocity versus midpoint depth, for example, or we can contour our well average or interval velocities perhaps contour them geostatistically using seismic processing velocities at distances far from the wells. Adding still more detail, we would like our model layer velocities to include variation with depth in some cases, because velocities often increase with greater degrees of compaction caused by thicker overburden (Figure 5c). For these situations we wish to have an instantaneous velocity data set to model, such as a time-depth curve from a vertical seismic profile, or check shot survey, or an integrated sonic log. This type of curve provides velocity variation over very small depth increments, hence instantaneous velocity. The simplest way to describe such variation is to model instantaneous velocity as a linear function of depth: V(z) = V0 + kZ, where V(z) is the instantaneous velocity at depth Z, and V0 and k are the intercept and slope of the line. Numerous other functions have also been proposed (Kaufman, 1953; Al-Chalabi, 1997b), some linear and some curvilinear. These functions are fit separately for each layer to ensure geological consistency. The authors advocate using the simplest model that fits the data acceptably well. So, interval velocity is used where appropriate (i.e., no consistent increase in velocity with depth), then preferably a linear model, and finally a curvilinear model only if necessary.
Continued on Page 17
16 CSEG Recorder November, 2001
ARTICLE
TRUE DEPTH CONVERSION: MORE THAN A PRETTY PICTURE
Continued from Page 16
Contd
instantaneous velocity with depth precisely - nor should it, because its purpose is not to describe the geology in that specific well, but rather the typical velocity within the geological unit overall. The goal is not to find a function that is an exact fit to the velocity vs. depth data for that layer for any one specific well; the goal is to find a specific parameter combination that produces a closer fit than any other combination for all wells, and that fits
Continued on Page 18
Figure 6
In practice no analytic function could represent the actual high frequency flutter of
November, 2001
CSEG Recorder
17
ARTICLE
Continued from Page 17
Contd
(4) from Al-Chalabi (1997a) where Vi and Ci denote the ith actual (observed) and function velocity values respectively, m is the number of sampled depth points, and q is the norm (q=2 in this case). In the crossplot space of the two free parameters (V0, k), the discrepancy values for each pairing are contoured. Each isodiscrepancy contour delimits a region in the parameter space inside which any (V0, k) combination produces a function that fits the well velocity data more closely than the value of that de-limiting contour. The area within an iso-discrepancy contour is an area of equally good parameter pairs. The discrepancy contour corresponds to a margin of tolerance. (Figure 7) Figure 8 If a single region of overlap can be found, then the reliability of the model is high since it applies to all wells used in the analysis. Thus, predicted values between the wells should be reliable. If the wells dont all overlap, but instead break into clusters (Figure 9), it may indicate that there are several different sub-areas within the overall area. These are often different fault blocks, or different facies associations. These situations can be handled by holding one parameter constant, such as k, and then solving for the other, allowing it to vary. Once calculated for all wells, it can then be mapped, providing a map of anticipated differences in uplift or facies.
Figure 7 There is no single parameter pairing that can be considered the exact solution, especially where a single well is concerned. A given parameter combination may, however, satisfy the data from more than one well. By making a composite discrepancy overlap plot of the discrepancy contours for the same layer in two wells (or three wells, or many wells), the region of overlap between the contours represents the (V 0,k) pairs that would produce a V(z) function that would fit both wells to within the appointed margin of tolerance. That is, any such parameter combination would provide a single function that applies to
Figure 9 The discussion thus far has focussed on calculating velocity models from wells. The patient reader has been waiting for the discussion to open up to the possibilities that seismic data offer to velocity modeling. The despairing reader may even have seismic but no well data. Are the benefits of velocity modeling still available in this situation? The following sections show that velocity modeling, including instantaneous velocity modeling, is still available even when only provelocities are available.
Continued on Page 19
18 CSEG Recorder November, 2001
ARTICLE
TRUE DEPTH CONVERSION: MORE THAN A PRETTY PICTURE
Continued from Page 18
Contd
Use all available velocity data to build a robust velocity model for depth conversion
While no-one would disagree with the advice, Use all available data, we must bear in mind that different types of data have different degrees of certainty, particularly well data versus seismic data. Well data can consist of vertical seismic profiles (VSP), check shot surveys, sonic logs, or some combination of these in several wells. VSP and check shots may be used directly, but sonic logs require corrections for drift to be comparable to a VSP or check shot survey in the same well (Reilly, 1993). Generally speaking, VSP are preferred most, then check shots, then integrated sonic logs, but the more wells available the better, even if it means mixing different types of time-depth curves. Well data are hard measures of depth - not completely without error, but the well depth measurements carry relatively low uncertainty. However, wells present us with velocity information that is spatially sparse, often clustered, and limited by well total depth. Further, well data overrepresent anomalous locations, such as structural highs. Seismic data offer a spatially dense, regular, and objective sampling, and cover the entire depth range evenly throughout the survey area. These traits offer the opportunity to overcome many of the limitations of using well data alone. However, seismic data are a measure of time rather than depth or velocity directly, and the provelocities derived from seismic are imaging velocities, not vertical propagation velocities such as in wells. Any effort that undertakes to combine hard (well) data (high certainty and low sampling density) and soft (seismic) data (low certainty and high sampling density) must honour the higher certainty of the well data. Geostatistics (spatial statistics) is the proper way to combine these two diverse types of data and retain proper weighting of well control, as well as to capture and maintain spatial trends (Chambers et al., 2000). For instance, kriging (which here is taken to include the various versions of kriging and cokriging) is a method of interpolation that uses specially-weighted combinations of data observed at known locations (such as wells) to predict unknown values at other locations. Kriging also provides estimates of the accuracy of the predicted values. The field of geostatistics presents many interesting techniques for integrating and mapping velocity, and analyzing spatial structures of velocity. We must restrict the scope of this short article to the topic at hand - depth conversion - and simply note that geostatistical analysis offers us tools to combine all available velocity data.
Continued on Page 20
November, 2001
CSEG Recorder
19
ARTICLE
Continued from Page 19
Contd
Summary
In this article we have touched on a number of issues with regards to translating seismic from time to true depth. Seismic imaging is a separate step and must be addressed before depth conversion. No depth conversion can correct for improper lateral positioning of events, because depth conversion is a vertical process only. Depth migration is currently the ultimate tool for lateral imaging, but it does not calibrate the seismic to true depth, because it does not use true vertical propagation velocities. This is not an error - imaging is a separate issue from true depth calibration. All imaging processes use a category of velocity that is more properly called provelocity, although imaging velocity or seismic velocity suffice as well. Provelocity is appropriate for imaging because seismic acquisition and processing involve both vertical and horizontal velocity to varying extents, but it is inappropriate for depth conversion - or depthing - because depthing requires strictly actual vertical propagation velocity (true velocity). True velocity is best obtained from vertical seismic profiles, check shot surveys, or calibrated sonic logs. Depthing can be done via a wide range of existing methods, too many to cover in any article, but which can be separated into two broad categories: 1) direct time-depth conversion, and 2) velocity modeling for time-depth conversion. Direct time-depth conversion ignores the structure (spatial patterns) of velocity, and operates at known depth points only (i.e., at wells) by forcing an exact or minimal error match between actual and predicted depths. Moreover, direct conversion only involves seismic times at well points - velocity information from seismic, and all the spatial benefits that go with it, cannot be used. Velocity modeling for time-depth conversion involves building a true velocity model using all available velocity data. This modeling may include various types of well velocities only, or calibrated provelocities only, or both. Modeling may use simple average velocity (single layer), or interval velocity (multi-layer), or instantaneous velocity (variation of velocity with depth). The goal is to determine a model that has some likelihood of working adequately between the known depth points, in addition to matching the known points. Some techniques can be used involving conditions other than final depth prediction accuracy, and can then be tested against the known points to determine their effectiveness. This is an independent way to predict depth because it uses velocity functions as the input rather than horizon depth and time at wells, and because it can involve provelocities in addition to or even instead of well velocities.
Figure 10 be amalgamated (averaged) into pseudo-wells to be used in instantaneous velocity function modeling just as the T-D curves from wells are used for instantaneous velocity modeling, albeit at a coarser time sampling (Figure 10). The averaging is used to smooth the error inherent in stacking velocity analysis. It can quickly be seen that even if we have seismic only and no well data or sparse wells, we can derive pseudo-wells and do instantaneous velocity modeling for our depth conversion. In this way we can use V(z) gradient functions to model velocity even if we are using seismically-derived velocities. In order to do this, though, seismic velocities need to be sufficiently detailed vertically to allow a robust V(z) curve to be derived. An approach to pre-stack velocity analysis has been developed to produce geologically consistent velocities in seismic processing (Crabtree, et al., 2000). In addition to providing a finer spatial sampling along the time axis, this technique forces a closer fit to well velocities and generally reduces the artifacts that are typically present in provelocities. Because of the greater areal coverage of seismic, one of the significant uses of pseudo-wells is to create many wells spread out across a study area and perform discrepancy contouring and overlap plots to look for clustering of pseudo-wells into areas of different velocity behavior. This points out areas of major facies changes or differences in uplift caused by faulting. The pseudo-well technique is a geological tool as well as a velocity modeling tool for time-depth-conversion. It can be used for the detection and evaluation of anomalouslypressured geological units, such as geopressured units (Gordon et al., 2000) that must be dealt with during drilling (or avoided), and basin-centered gas accumulations (Surdam, 1997).
Continued on Page 22
20 CSEG Recorder November, 2001
ARTICLE
Continued from Page 20
Contd
EDWARD L. ETRIS Edward (Ned) Etris received M.Sc. (1987) and Ph.D. (1991) degrees in geology from the University of South Carolina, specializing in quantitative sedimentology. Ned is currently Manager of Geology at Scott Pickford Canada. He is an experienced reservoir characterization geologist and numerical reservoir model developer, with particular strengths in the evaluation and integration of core, log, and seismic data, including statistical and geostatistical analysis. He came to Canada in 1991 to work for Canadian Hunter Exploration Ltd., then later Ranchmens Resources Ltd. and Crestar Energy Inc., gaining a wide range of production company experience in Western Canada. His work has ranged from research to development drilling and pool exploitation, and from regional prospecting to detailed reservoir characterization. Since joining Scott Pickford in 1999 he has directed work on time-depth conversion, 3D geological modeling, petrophysical evaluations, and tight gas sand evaluations, and played a key role in integrated geophysical-geological-engineering studies within Scott Pickfords Reservoir Modeling Group. Ned has numerous papers and professional presentations to his name, and has taught industry courses on geostatistics.
NICK J. CRABTREE Nick Crabtree received an M.A. (Hons) Natural Sciences (Geological Sciences) degree in 1992 from Jesus College, Cambridge, UK. (The college, part of the University of Cambridge, was founded in 1496!) Nick is currently Technical Manager of Depthing Research and Services, at Scott Pickfords Croydon UK office. He has worked on the Velocity Modeling Using Statistical Analysis of Seismically Derived Velocities project; served as technical lead on the development of VELIT, Scott Pickfords velocity modeling software; and presented numerous papers at the SEG, CSEG, and EAGE. In 1999, after a year in snowy Calgary, Nick returned to the Croydon office to head up the RISKIT research project studying uncertainty and sensitivity analysis of velocity modeling and depth conversion. He provides technical support and mentoring in the areas of depth conversion, velocity modeling and reservoir characterization. Nicks professional interests also include seismic attribute analysis, reservoir characterization, integration, and mapping.
JAN DEWAR Jan Dewar graduated from the University of Alberta in 1981 with a B.Sc. in Physics. Jan is currently working with Scott Pickford in Calgary, with a special enthusiasm for communicating technical concepts including AVO, Inversion, Modeling, VSP and Transfer Filter processing, Rock Physics, and just about anything else that can be puzzling to the average bear. jdewar@scopica.com
22
CSEG Recorder
November, 2001