Projection before picking: a statistical approach to tomographic imaging

David C. Henley

The goal of tomographic imaging with seismic data is to create a velocity image of a portion of the earth from a set of transit time values for some seismic event, as measured along a multitude of diverse paths through the target medium. These transit time values, one for each seismic path, are subject to various estimation errors, which can be quite large in the case of miss-identified seismic events. Since the process of picking the transit times is usually fraught with error, we propose an imaging method wherein the transit time variable is first associated with a unique probability density function for each input trace, a 3-D volume of probability-weighted transit times is created, and the volume is back-projected to yield a probability-weighted image volume which is less noisy than the raw data volume. Each surface of constant probability within the image volume constitutes a potential tomographic transit time image. Interpretation, in this context, then becomes the extraction, by one of several possible processes, of an image surface from the back-projected volume. Hence, by interchanging the processes of back-projection and picking, we avoid explicit miss-picks on noisy data.