DERIVING RELIABLE GEOLOGICAL INFORMATION FROM SEISMIC TOMOGRAPHY

Three-dimensional models of seismic wave speeds within even small volumes in the Earth require huge numbers of parameters to specify them. Realistic seismic data sets almost never contain comparable amounts of information. To circumvent this problem, it is conventional to incorporate “regularization” constraints into seismic tomography methods, for example requiring derived models to be close, in a least-squares sense, to an arbitrary starting guess. Such added constraints may have little or no observational basis, and introducing them can introduce artificial features, such as regions of low or high wave speed, that are not required by the observations. These artifacts may then lead to baseless geological interpretations.

To avoid such difficulties, Tarantola (Nature Physics, v. 2, pp. 492-494, 2006) suggested that seismic tomography practice be altered so as not to seek “optimal” models but rather to test specific scientific hypotheses.

We present examples, using both synthetic and real data, of artifacts that conventional seismic tomography methods can introduce, and show how they can be avoided using a hypothesis-testing approach. In particular, we show examples of “four-dimensional” seismic tomography, testing the reality of apparent temporal changes in seismic wave speeds in volcanic areas and in both natural and exploited geothermal areas.