Abstract: We demonstrate a method for 3-D subsurface tomography in which crosswell tomographic signals are inverted in a novel way to yield 3-D parameter estimates. The inversion is accomplished via an approximate extended Kalman filter, in which geostatistics are used to implicitly smooth and regularize the inversion. Cluster analysis is used to dynamically determine the parameter dimensionality and structure; the ultimate resolution of heterogeneity is controlled by a cluster tolerance criterion. The size of the domain is incrementally increased as the estimation progresses, which keeps the computational requirements manageable. The method ultimately estimates the number, geometry, values and covariance of parameters.
[edit database entry]