The Age of Migration

[Note: A version of this blog entry will appear in World Oil (July, 2010)]

Seismic data processing plays a key role in exploration. In the modern search for hydrocarbons, few wells are drilled with seismic data. The role of seismic is reduction of risk; risk of drilling dry holes, marginal producers, or getting reserve estimates seriously wrong. It seems appropriate to pause in 2010 and consider progress in seismic data processing over, say, the last 30 years. It is a big subject, the annual SEG meeting alone generates over 1100 expanded abstracts, the vast majority on seismic topics. We will touch here on a couple of first-order advances (migration and anisotropy) that have changed the way seismic work is done around the world every day.

First we should define seismic data processing. It is a vast and growing arsenal of computational techniques that attempt to remove wave propagation effects or noise in order to create an image of the subsurface. A popular free software package called SeismicUn*x, for example, consists of over 300 individual programs for the processing, manipulation, and display of seismic data. Importantly, seismic data processing does not include calculations like attributes, AVO, and inversion that mine the data for further information. These methods are important, but not properly part of our subject.

By 1982 the broad outlines of 2D prestack migration were taking firm shape. 2D may seem quaint today, but it was a massive strain on computer power. Furthermore, prestack migration was already known to be very sensitive to velocity errors, meaning not one, but many, passes of prestack migration were needed.

One approach to this difficulty was decoupling of prestack migration into separate steps. In this view, prestack migration was sequential application of normal moveout (NMO), a mystery process, common midpoint stacking, and finally poststack migration. This mystery process carried the burden of making the decoupled processing flow give precisely the same result as prestack migration. It came to be called dip moveout or DMO.

A working commercial implementation was available by 1978 and DMO was in general use by 1984. Over the next two decades, it was extended to include velocity variation (vertical and lateral), mode conversion, anisotropy, and 3D. Some migration purists see DMO as a trick; an annoying distraction of resources and effort that should have gone into the real problem of prestack migration. Perhaps. But DMO served a definite purpose in bridging the gap between 1980s computer power and the computational needs of prestack migration.

Seismic migration had a long history before this. Lateral positioning errors were understood almost as soon as people started shooting data. There was important work in the 1950s, and in the early 1970s a comprehensive view evolved of what migration was and how to do it. Then the great foundation papers came in 1978 – Kirchhoff, phase shift, and F-K migration – all in a single issue of GEOPHYSICS. Seismic migration was a mature science by 1985.

From the late 1980s through today, we have seen the age of prestack depth migration. The need for this technology arose from the failure of standard processing. The flow of NMO, DMO, common midpoint stack, and post- stack migration, gives a satisfactory image when the earth is simple. But exploration was pressing into deeper water, snooping around salt overhangs, testing subsalt rock formations, dealing with extreme topography, and trying to image overthrust areas. All these cases involve significant 3D lateral velocity variation. When things get tough enough everything is a migration problem. Under these conditions the decoupled processing flow simply fails to provide a geologically meaningful image and prestack depth migration is needed.

Standard processes are ultimately based on constant velocity physics or perhaps vertical variation. In a continuum of progress the migrators have incorporated more and more physics into the migration process – anisotropy, 3D, and strong lateral velocity variations. There has been remarkable theoretical progress in finding new ways to implement migration; various Fourier methods, finite difference (including reverse time), gaussian beams, screen propagators, and the venerable time-space domain Kirchhoff migration.

Even today Kirchhoff depth migration is the dominant technique, perhaps because of its superior ability to accommodate arbitrary coordinates for each source and receiver. Increasingly during the late 1980’s and early 1990’s the burden of travel time computation was spun off from Kirchhoff migration and sequestered in ray tracers of every increasing complexity. Radiation patterns, attenuation, various elastic wave phenomena, anisotropy, multi-pathing, and whatever else we think is important can ultimately be jammed into ray tracing. The migration program itself begins to look more and more like a database matching up coordinates, travel times, and amplitudes.

A bit more about anisotropy is in order. The tendency of seismic waves to have directional velocity was well-known fifty years ago and the theory was worked out in detail fifty years before that. But anisotropy was profoundly ignored in seismic data processing until the late 1980’s. Why? In standard land shooting we measure only the vertical component of motion then do everything we can to supress shear waves, and anisotropy is an elastic effect that was primarily studied in relation to shear waves. It seemed unlikely you could do anything to estimate it or remove it’s effects without measuring three component data to observe the full elastic wave field. This all changed in 1986 when it was shown that anisotropy influences standard P-wave data. A major theme in migration since then has been inclusion of ever more complex anisotropy.

Once you look for anisotropy it is everywhere, due to shale, fractures, thin layering, regional stress, etc. The important thing is this: If the subsurface is significantly anisotropic and you do isotropic data processing the image is degraded – amplitudes are wrong, reflector segments are not at the correct depth or lateral position, fault terminations are smeared, and so on. The interpreter is compromised by the processors’ isotropic worldview.

The last 30 years deserve to be remembered as the foundation age of seismic migration. Progress will continue, but the foundation is only built once.