UTIG RESEARCH PROJECTS ARCHIVE

Principal Investigator: Mrinal K. Sen

Funded by: Texas Higher Education Coordinating Board

Inversion of geophysical data to obtain subsurface material properties is critical to increasing our quantitative understanding of geologic structures and lithology and has direct application in hydrocarbon exploration and exploitation. Seismic waveform inversion seeks to determine compressional wave velocity, Poisson's ratio and impedance contrasts by minimizing the differences between observed and synthetic seismic data based on proposed subsurface structures and a particular type of wave propagation. This problem is nonlinear and can be solved using nonlinear optimization procedures such as simulated annealing (SA) and genetic algorithms (GA). However, depending on the seismic data actually used and the a priori information available, seismic waveform inversion can be made approximately linear so that solution by iterative linear approaches is possible. Experience with iterative linear and nonlinear invesre methods shows that each approach has its unique merits and disadvantages. For example, nonlinear methods, because of their high computational cost, necessitate a sparse parameterization of the subsurface material properties. This low resolution approximation to the subsurface, however, can be derived with little a priori knowledge and the posterior probability density and covariance can be estimated. Linear techniques apparently achieve higher resolution results but require a starting model that is close to the subsurface structure. These results are relevant therefore only in the context of this starting model. We believe that elements of nonlinear optimization and iterative linear methods can be combined to achieve two important goals: increase the rate of convergence and thereby lower the computational cost; and, improve resolustion without significantly increasing the computational burden. If these objectives can be achieved, seismic waveform inversion could be routinely applied, thereby, significantly improving our estimates of material properties.