Deepwater geohazard prediction using prestack inversion of large offset P-wave data and rock model N. C. DUTTA, WesternGeco, Houston, Texas, U.S. S uccessful deepwater drilling is defined as reaching the total well depth in the desired hole size safely (i.e., not exceeding the fracture pressure) while controlling hydrocarbon or water influx and ensuring that casings are placed across desired intervals. The oil and gas industry has succeeded in this mission thousands of time around the world but drilling challenges abound, especially in deepwater. These challenges include: • Smaller tolerance between pore pressure and fracture pressure causing narrow pressure margins while drilling (What is the proper mud weight window?); • Shallow-water flow hazards due to pressured aquifer sands (How do you predict occurrence before spudding a well?); • Excessive casing programs/small hole at total depth or unable to reach total depth (How do you design a proper casing program i.e., depths and sizes?); • Slow rate of penetration and incorrect use of bit types combined with excessive bit wear (How do you minimize rig time without sacrificing safety?). The industry has successfully mitigated many of these hazards but more needs to be done when planning a well to ensure that all appropriate and available technologies are used properly and effectively. A case in point is the losses due to shallow-water flow (SWF) hazards in deepwater. A 1999 conference revealed that, although the industry was spending about US$1.6 million per well in dealing with problems associated with SWF, only about 20% of this amount was for predrill prediction. The rest was for remediation after SWFs were encountered. Seismic techniques for geohazard prediction (pore pressure and SWF) have come a long way since the pioneering work of Pennebaker (1968) and Reynolds (1970) who used seismic stacking velocities for pressure prediction, based on the fundamental analysis of overpressuring in clastics basins by Hubbert and Rubey (1959). Great advances in the last decade in 3-D seismic acquisition and processing, especially in imaging and inversion technologies, allow geohazard prediction with more certainty and clarity. Current investigations suggest that geohazard formations and their properties (i.e., velocity, density, and rigidity) are amenable to prediction methodologies that use constrained inversion of large-offset conventional P-wave data in conjunction with rock physics models. The key steps in pressure prediction using seismic (Figure 1) involve: • Assessing the cause of overpressure (compaction, for example); • Developing a rock model for relating rock attributes (velocity, porosity, Poisson’s ratio, for example) to higher than normal pore pressure or lower than normal effective or differential pressure (which is the stress acting on the rock matrix); and Figure 1. A typical flow chart showing pressure prediction methodology and highlighting the major steps in the work flow. Figure 2. Shale burial diagenesis and its effect on the density and slowness of Gulf Coast shales. (a) shows the smectite to illite conversion in a well from the Gulf Coast (Freed 1982) created by the burial history simulation. (b) shows how density of the shales is altered when plotted against slowness (inverse of velocity) as obtained from wireline logs. • Using a fit-for-purpose seismic velocity that resembles rock velocity and not processing velocity as stated very clearly by Al-Chalabi (1994). Role of temperature on compaction. Although there are multitudes of causes of geopressure, undercompaction or compaction disequilibrium is thought the most common in clastic basins. This is the most dominant mechanism in deepwater. However, compaction has often been misrepresented in the literature. It is not just reduction of porosFEBRUARY 2002 THE LEADING EDGE 193 Figure 3. Bulk density versus slowness of shales from a deepwater well in the Gulf of Mexico. The color codes indicate depth. Figure 4. Trends of Poisson’s ratio of water and gas saturated unconsolidated sands obtained from the work of Prasad et al. (2000) at Stanford University. ity and expulsion of pore water due to increasing overburden load. Burial history plays a key role—i.e., the effect of temperature and time on the sediment compaction properties, especially that of clays in shales (Dutta, 1983). Clays undergo burial metamorphism as they undergo compaction (for example, the smectite-to-illite transformation within the mixed-layer clay systems in the Gulf of Mexico). The phenomenon causes reordering of clay platelets and redistribution of effective stress and strongly depends on the time-temperature history of the sediment. This implies that compaction is both time and temperature dependent and that the notion of rocks compacting along a single “virgin” compaction curve (assuming that porosity is a function of effective stress only) is invalid. Porosity is a function of effective stress, the temperature that it was subjected to as it continued along a burial path, and the time that the rock spent in a certain temperature window. The mechanism is kinetic in nature, similar to 194 THE LEADING EDGE FEBRUARY 2002 kerogen conversion leading to oil generation. This also suggests that any approach that uses a single compaction curve to describe geopressure may be questioned. Recently, Lahann (1999) and others discussed the impact of smectite diagenesis on compaction profiles and compaction disequilibrium and suggested that two limiting compaction curves—one for smectite, one for illite—bound the compaction profiles. His analyses were based on wireline log and in-situ pressure data. My internal studies agree with Lahann’s conclusions. Figure 2a shows an imprint of smectite diagenesis on x-ray diffraction of shales from a well in the Gulf Coast (DOE 1). The x-ray measurements were made on fine grain sediments (less than 5 micron) by Freed (1982). The burial metamorphic process is kinetic in nature—namely, both temperature and geologic time control the process. Figure 2b shows a crossplot of petrophysical properties of shales from the same well (wireline log measurements)—namely, bulk density and slowness (inverse of velocity). The rocks are Miocene age. The figure suggests that diagenesis causes an increase in bulk density for a given transit time or slowness. Similar behavior is also noticed in a deepwater well from the Gulf of Mexico (Figure 3). This suggests that wellknown relationships that relate velocity to bulk density but ignore the effect of temperature (and time) are not valid for geopressure estimation which uses velocity data to extract porosity and then relates this porosity to effective stress to compute pore pressure. The current investigation uses a temperature-dependent compaction behavior. Porosity is a function of the clay type, effective stress, time, and temperature. Note that the direct effect of increasing temperature on the velocity of a fluid-filled porous rock is negligible; the effect on the porosity and the effective stress is not. Poisson’s ratio of geopressured rocks. Poor grain-to-grain contacts in overpressured zones with low effective stress affect the S-velocity more than the P-velocity. This has been observed in the laboratory (Domenico, 1984) and modeled (Xu and Keys, 1999). VP/VS (which is directly related to Poisson’s ratio) is a key parameter that relates to low effective pressure or high pore pressure, especially under SWF conditions (pore pressure close to fracture pressure). This has not received much consideration in pore-pressure prediction methodologies because the traditional seismic approach for pressure prediction used conventional P-velocity data only. Figure 4 plots Poisson’s ratio versus effective stress based on data from the Stanford Rock Physics laboratory for two cases, sand filled with water and sand filled with gas. Note the behavior of Poisson’s ratio at low effective stress. For the water-filled case, the ratio increases with increasing pressure (lowering effective stress). The opposite is true for the gas-filled case. This suggests a very different seismic signature for the two cases in the AVO domain. The present investigation exploits this fact by inverting CMP gathers using full waveform inversion that allows extraction of VP/VS from large offset P-wave data. Miley and Kessinger (1999) demonstrated how to use this effect on shear velocities determined from mode-converted wave reflections for subsalt pore-pressure prediction. Trend-line versus effective-stress approach. It has been common to predict pore pressure before drilling from conventional seismic stacking velocities with a normal compaction trend analysis using the well-known Eaton approach (Eaton, 1972). Velocities that appear to be slower propagation velocities” (Al-Chalabi, 1994). Second, these velocities lack resolution in depth and may not be very useful for drilling applications. Third, in deepwater, sediment loading has often been so fast that pressures in these sediments are above the hydrostatic pressure right below the mud line (rocks are geopressured). This prevents development of a normal compaction trend. Even when a velocity trend is constructed (with depth as a variable) by curve-fitting pressure data from an analog well, these trends often violate the fundamental bounds of rock physics. The velocity neither approaches the correct sediment velocity just below mud line nor approaches a limFigure 5. High frequency velocity, bulk density, and Poisson’s ratio iting velocity appropriate for very low obtained from the prestack full waveform inversion of a CMP gather porosity rocks at very high overburden shown on the left. The uncertainties are shown in yellow. pressure (Dutta, 1997). Approaches based on the effective-stress concept are always preferred to other approaches that use depth as a variable in describing compaction behavior. Pore pressure is defined as the difference between overburden stress and effective stress. Effective stress affects the grain-to-grain contacts of clastic, sedimentary rock, and consequently, the velocities of seismic waves propagating through such rock. The rock model has various components: relations between porosity, lithology and velocity, clay dehydration, and transformations relating both density and Poisson’s ratios of the sediments to effective stresses acting on the matrix frame. The key inputs that drive the rock model are velocities (P and S) obtained from a variety of velocity tools. Iterative velocity analysis, calibration, and interpretation are key and essential steps to ensure that the velocity fields are within the realm of expected rock velocities. Figure 6. Pore pressure in 3-D from analysis of every CMP and every time sample of a 3-D data set covering ~ 650 km2 using the generalized velocity inversion technique of Mao et al. (2000). Figure 7. Pore pressure analysis using velocities from tomographic and prestack depth migration analysis. These velocities were conditioned using constraints based on rock physics and those discussed by AlChalabi (1994). than normal compaction velocities are indicative of overpressure, which then is quantified using an empirical equation. There are several problems with this approach. First, conventional seismic stacking velocities are totally unsuitable for pressure prediction because they are not “rock or Velocities from inversion of prestack gathers (traveltime and amplitude). Velocities from seismic data are the key input to any seismic-based pressure prediction methodology. They are derived from inversion of prestack seismic data, from either traveltime or amplitudes or some combination of both. Although there is a myriad of seismic velocities, most are not fit for pressure analysis. The main reason has been discussed above. Most of these velocities resemble some sort of processing or imaging velocities and not rock velocity. Further, they must be subjected to constraints based on the basic principles of rock physics, such as the close relationship between VP and VS and that between VP and density for a given rock type. In a recent paper, I discussed various seismic velocity tools that can be used for pore-pressure evaluation and how to impose rock physics constraints on velocity analysis. Some commonly used velocity tools that have had considerable success for pore-pressure imaging (PPI) are: • Semblance velocity analysis (based on Dix-inversion) • Horizon-based velocity analysis at every CMP and at every time sample • Tomography and prestack depth migration • Velocities from poststack inversion • Velocities from prestack full-waveform inversion Most success in PPI is achieved using a fit-for-purpose toolkit for velocity analysis. The goal is to derive velocities close to rock velocities—i.e., velocities that are related to effective stress and pore pressure. The key steps in the FEBRUARY 2002 THE LEADING EDGE 195 Comparison of observed and synthetic angle gathers tions. Further, prestack full waveform inversion provides estimates of shear velocity in addition to a high resolution P-velocity. The genetic algorithm (GA) inversion methodology discussed by Mallick (1999) is an optimization algorithm used to find a suitable P- and Svelocity earth model by minimizing the misfit between observed angle gathers and synthetic ones. The process is very computer intensive; however, it is the only method that yields density and Poisson’s ratio from prestack data. AVO techniques merely approximate these variables within the confines of the linear inversion scheme. Density is used to estimate overburden pressure, whereas Poisson’s ratio is used for fracture pressure computation and shallow water flow hazard identification (Dutta, 2001). Figure 8. A comparison of the synthetic seismogram (shown on the right) The prestack inversion begins with created using the GA algorithm as discussed by Mallick (1999) with the an automatic NMO-correction, yielding observed gather shown on the left. The data are from a deepwater well in an initial rms velocity. Next, interval the Gulf of Mexico. velocities are derived from rms velocities and a background density is determined as a function of depth from the interval velocity field. In addition to the moveout information, the GA inversion then uses AVO data from the NMO-corrected prestack gathers to determine initial density and Poisson’s ratio values. Once the initial model is selected, GAgenerates random earth models (full waveform synthetic seismic), say at every 20 ms within certain search bounds. It then uses standard GAsearch algorithms to find the optimum earth model—the one that generates synthetic seismic data, which matches the input seismic data using an exact wave equation based on forward modeling. This exact modeling method calculates all primary, multiple, and mode-converted reflections. Consequently, all interference or tuning effects due to thin layers and transmission effects due to velocity gradients are accurately modeled in the inversion. Because GA uses a statistical optimization process to search for the elastic model, it also generates a measure of Figure 9. A comparison of the inverted P-velocity and Poisson’s ratio from the uncertainties in estimation of the elasthe GA inversion with those from the actual well logs for the CMP data tic model as an additional output. shown in Figure 8. The uncertainties are shown in yellow. Figure 5 shows an example of the VP, density, and Poisson’s ratio derived from process involve velocity analysis that includes residual this process. Note the high frequency nature of the velocNMO correction, higher order moveout effects, analysis ities predicted from the prestack inversion process. These of velocity anomalies, accounting for raypath bending, velocities are more appropriate for drilling applications anisotropy correction, and velocity interpretation to honor than those based on conventional velocity analysis. the local geology. Extracting velocities from prestack fullHowever, inversion is inherently nonunique. Two key waveform inversion is an emerging technology that is parsteps minimize problems associated with nonuniqueness— ticularly suited for drilling applications. using the largest feasible offset data and imposing rockphysics-based constraints on the inversion. A distinct Prestack full-waveform inversion. Most velocity analyadvantage of this method is that it is not as severely depensis tools mentioned above provide velocities in the lowdent on the quality of the reflection data as AVO inversion frequency regime and they may not be useful for drilling or poststack inversion. Because the GA procedure allows applications where mud programs and casing point locafor “hidden layers” in the inversion scheme, it can identions require more resolution. In this regard, full-waveform tify pressure cells for which no associated reflections exist. prestack inversions of velocities from seismic amplitudes are very useful and especially suited for drilling applica196 THE LEADING EDGE FEBRUARY 2002 ramping is over 2 ppg. Because of the complexity of geology, tomographic inversion and depth migration were essential in this case. But, although prestack time migration revealed the pressure ramping, the magnitude of pressure and the depth at which point the ramping started were wrong. A casing program based on the prestack time migration would have resulted in unsafe drilling conditions. Figures 8-11 show applications of prestack and full waveform inversion (GA procedure). Figure 8 shows a single CMP gather (angle gather) at a deepwater well (water depth greater than 5000 ft) obtained from the GA inversion prior to drilling. Agreement between the GAFigure 10. Results of pressure prediction for a well in real time using the GA derived result and the actual CMP angle inversion technique and its comparison with the actual pressure data (RFT) gather is good. Figure 9 compares preobtained after drilling the well. The pressure profile is appropriate for the dicted velocity logs (P and S) to those inversion and the velocity results shown in Figures 8 and 9, respectively. obtained after drilling. Overall agreement between pseudologs and actual logs is striking. A comparison of the predicted pressure profile using VP from the GA inversion with those from drilling data (Figures 10-11) indicates that predicted pressures agree with measured values to within 0.5 ppg. Prestack inversion revealed pressure ramping at the deeper depth. This was a concern prior to drilling. The low-frequency model based on velocity analysis did not indicate its presence; it happened in a lithologic column devoid of reflections (shaly interval). Check shot data prior to ramping allowed calibration of velocity data from the inversion. Then a look-ahead GA inversion provided correct velocities ahead of the bit that can be used to predict pressure profile while drilling. Planned downtime on the rig due to casFigure 11. A comparison of the predicted versus the actual pressure data in ing provided an ideal opportunity for the same well for which prestack full waveform results were presented in this evaluation. The results increased the Figures 8-10. confidence of the drillers, provided a correct casing location, and eliminated an Examples of pressure images from seismic inversion. extra casing and provided a proper mud weight window Figures 6-7 show pore-pressure images using low-frefor drilling ahead without exceeding fracture pressure. quency velocity models discussed above. Figure 6 is an In our example, we observed the influence of effective example of semblance-based velocity analysis on 3-D data. stresses on the shear modulus in-situ. Figure 12 shows this It analyzes velocities along interpreted geologic horizons modulus determined from log measurements as a funcfor every CMP and at every time sample. It does not use tion of depth and shale content (gray scale). Looking at the conventional Dix-model, but a generalized inversion the (light colored) shales in the shallower two thirds of the process that computes interval velocities from rms velocdata, we observe a continuous increase of the shear modity functions (Mao et al., 2000). The figure shows a presulus with depth, indicating progressive compaction. sure cube in 3-D. The velocity analysis enables us to see Deeper in the zone where the pressure ramp is encounthe pressure image in 3-D in significantly greater clarity tered, the abnormally low effective stress leads to a prothan with other methods. High-pressure pockets (presnounced decrease in the shear modulus. The 15% increase sure approximately 16 ppg) are clearly visible. of the pore-pressure gradient in this zone leads to an Figure 7 is an example of the use of tomography and approximately 25% decrease in the shale shear modulus prestack depth migration for pressure analysis. Before over the same interval. This is remarkable and suggests depth migration, velocity model building utilized reflecthat both shear-wave data and very large offset P-wave tion (grid) tomography that accounts for raypath bending. data may be very good indicators of high pore pressure. The process used traveltime computation and several iterIt also suggests that Poisson’s ratio versus depth predications to construct velocities consistent with local geology tion using prestack inversion could be a powerful tool to and rock physics constraints. This figure dramatically estimate this important attribute prior to drilling—pershows a clear transition from low to high pressure. Pressure haps more meaningful than those obtained using conFEBRUARY 2002 THE LEADING EDGE 197 computed pressure profiles. These uncertainties result from very important factors (velocities, rock parameters relating velocities to effective stress, and compaction trend analyses) that need to be considered. Third, and perhaps most challenging, is the cultural gap between the seismic community (the provider of the technology) and the drilling community (the user). Much remains to be done. As a practitioner for more than 15 years of the seismic technologies discussed here and having applied these pressure estimations in basins throughout the world, I would like to close this article with a very positive note. The future is bright. The availability and ever declining cost of high-speed computers will make it possible to invert 3-D seismic volumes in the prestack domain, using full waveform inversion technologies, to obtain pressure images with unprecedented detail and clarity. The evolution of this technology will be like the advent and use of 3-D prestack depth imaging. Figure 12. Shear modulus versus depth for a deepwater well in the Gulf of Mexico. The data were obtained using shear velocity and density logs. ventional AVO (small offset, linear inversion). Conclusions. This work suggests that formation pore pressure could be predicted from seismic data provided care is taken to condition the seismic velocities and ensure that rock velocities are extracted from seismic velocities. The analysis and the examples show that the largest feasible offset data set should be used for PPI. A fit-for-purpose velocity tool should also be used. Tomography and prestack depth migration should be used when significant imaging problems, lateral velocity gradients, and raypath bending effects are encountered. In most cases, prestack time migration may suffice. Further, the rock model relating velocity to effective stress must include geologic variables relevant to that basin. An example is the role of temperature in the Clastic Tertiary Basin of the Gulf of Mexico. Prestack full waveform inversion yields a pore pressure versus depth profile with more vertical resolution than possible from conventional velocity analysis. Further, the resulting Poisson’s ratios can be used for fracture gradient and lithology prediction. High-frequency pressure variations from inversion make it an attractive technology for drilling applications. Nonetheless, the inherent nonuniqueness of inversion results requires imposing rock physics constraints. Are there challenges ahead? Indeed. First, an understanding of rock physics is a must for a reliable prediction of pore pressure. Unfortunately, the industry relies heavily on empirical formulations in this area. Such approaches may be defended in a given area but do not allow knowledge transfer from one basin to the other. Second, many practitioners routinely ignore uncertainty estimates on 198 THE LEADING EDGE FEBRUARY 2002 Suggested reading. “Seismic velocity—A critique,” by AlChalabi (First Break, 1994). “Shale compaction and abnormal pore pressures,” by Dutta (SEG 1983 Expanded Abstracts). “Pressure prediction from seismic data: Implications for seal distribution and hydrocarbon exploitation in deepwater Gulf of Mexico,” by Dutta (Norwegian Society of Petroleum Special Publication No. 7, edited by Moller-Pederson and Koestler, Elsevier Press). “Graphical method predicts pressure worldwide,” by Eaton (World Oil, 1972). “Some practical aspects of prestack waveform inversion,” by Mallick (GEOPHYSICS, 1999). “Rock lithology and porosity determination from shear and compressional wave velocity,” by Domenico (GEOPHYSICS, 1984). “Overpressure prediction using converted mode reflections from base of salt,” by Miley et al (SEG 1999 Expanded Abstracts). “Study of coupled effect of pressure, frequency, and fluid content on P- and S-velocities,” by Xu et al ( SEG 1999 Expanded Abstracts). “Impact of Smectite diagenesis on compaction equilibrium,” by Lahann (1999 Del Lago Geopressure conference proceedings). “Clay diagenesis and abnormally high fluid pressure,” by Freed (SEG 1982 Expanded Abstracts). “ Spatially continuous velocity analysis and automated velocity model building,” by Mao et al (SEG 2000 Expanded Abstracts). LE Nader Dutta has a PhD in physics. He joined the oil industry about 26 years ago with Shell Oil. In 1986, he joined ARCO’s Technology Center to direct the Geoseismic Interpretation Group and three years later joined the Deepwater Group of BP in Houston. Dutta joined Baker Hughes’ Inteq Division as senior science advisor in 1999. Dutta was Strategic Business Development Manager and Worldwide Operations’ Manager of Lithology, Fluid and Pressure Imaging area in WesternGeco’s Seismic Reservoir Services Division (WesternGeco is a joint venture between Schlumberger and Baker Hughes). On 1 November 2001, he was appointed chief geoscientist at WesternGeco with responsibility for all technology-related activities underlying seismic reservoir services. Dutta has worked in various aspects of seismic wave propagation, including borehole geophysics, rock physics, basin modeling, seismic while drilling, and pore-pressure technology. Corresponding author: Ndutta@houston.westerngeco.slb.com