Ocean acoustic tomography: Integral data and ocean models

Tomographic data differ from most other oceanographic data because their sampling and information content are localized better in spectral space than in physical space. Approximate data assimilation methods optimized for localized physical space measurements can lose much of the nonlocal tomographic information, degrading the potential performance of the tomographic data in fixing the model state. In addition, methods which require inverting the tomographic data to a physical space,arid before insertion into the model as point measurements incur special problems because of the nonlocal nature of the errors. We give a simple one-dimensional example to illustrate the problems that can arise. Fortunately, methods that directly assimilate integral measurements and preserve all of the information in the tomographic data, such as the Kalman filter, are becoming more practical as computer speed grows.