NONUNIQUENESS IN TRAVEL-TIME TOMOGRAPHY - ENSEMBLE INFERENCE AND CLUSTER-ANALYSIS

We examine the nonlinear aspects of seismic traveltime tomography. Thi s is accomplished by completing an extensive set of conjugate gradient inversions on a parallel virtual machine, with each initiated by a di fferent starting model. The goal is an exploratory analysis of a set o f conjugate gradient solutions to the traveltime tomography problem. W e find that distinct local minima are generated when prior constraints are imposed on traveltime tomographic inverse problems. Methods from cluster analysis determine the number and location of the isolated sol utions to the traveltime tomography problem. We apply the cluster anal ysis techniques to a cross-borehole traveltime data set gathered at th e Gypsy Pilot Site in Pawnee County, Oklahoma. We find that the 1075 f inal models, satisfying the traveltime data and a model norm penalty, form up to 61 separate solutions. All solutions appear to contain a Ce ntral low velocity zone bounded above and below by higher velocity lay ers. Such a structure agrees with well-logs, hydrological well tests, and a previous seismic inversion.