In ocean acoustic tomography, the travel time along a ray path is a weighted average of the propagation speed along the path. Because oceanographers generally have intuition about point measurements or simple averages, it has been necessary to transform travel time data into point values (maps) before communicating the results. The transformation can be done with a variety of methods, ranging from exhaustive Monte Carlo searches to Backus-Gilbert constrained estimation. The transformation converts travel time data with more or less independent errors to point value estimates with correlated errors that may have complicated, nonlocal structure. Since the error bars usually presented with an ocean map do not include the correlations, they do not accurately reflect the information content of a tomographic dataset. In addition, it is no longer possible to distinguish between the data errors and sampling blind spots by examining the error bars (or even the error covariances). Communicating tomographic results thus requires more effort, and more plots. Resolution (or averaging) kernels show how the estimate at a point is a weighted average of the entire field (with the average becoming more local as the number of rays increases). Null space vectors show fields that may be added to the estimated map without changing the data significantly. Given that the goal of ocean observations is to test dynamical hypotheses, it is also reasonable to consider transforming hypotheses into constraints on the travel times, rather than transforming the travel times into constraints on physical space hypotheses. [Work supported by ONR and ONT.]
Read full abstract