GLOBAL OCEAN DATA ASSIMILATION
Some recent results from the US effort can be viewed at
"http://puddle.mit.edu/〜detlef/OSE/global.html". In this particular case, a globai ocean circulation model has been made consistent with six years of altimetric data, monthly Reynolds sea surface temperature, NCEP meteorological surface boundary conditions, and monthly Levitus climatology. The optimization was carried out using the so-called adjoint model method, also known as 4-D-Var in meteorology, or as the Pontryagin principle in the engineering literature. Another example is the work reported by Fukumori et al. (1999) who used an approximate Kalman filter and smoother. These results are preliminary, and many details remain to be worked out. Nevertheless, the results demonstrate that advanced optimization methods can be applied to the ocean circulation problem. Given concerted observational and modeling efforts, it should be possible to obtain vastly improved estimates of ocean circulation within the next five to ten years.
We expect that if such calculations became routinely available and continually improved, in the way that numerical weather analyses have improved over time, then many facets of oceanography and of other fields would come to rely upon the resulting products. For example, improved circulation estimates would be used for testing theory and models of the general circulation, for physical-biological and bio-geochemical modeling, for studying air-sea interaction, for regional and coastal applications, for fisheries, exploration, navigation, etc. Twenty years following the launch of SEASAT by NASA, satellite oceanography is well established and it provides surface boundary conditions for the circulation with global coverage and at reasonable cost. Circulation models are also vastly improved compared to twenty years ago, especially due to increased computer capability and improved parameterizations. What is still missing are continuous measurements within the water column to complement the observations from space.
The problem of lack of measurements in the interior can be illustrated using, for example, results from the numerical study of Menemenlis and Chechelnitsky(1999). In this study, artificial errors with known statistics were added to the large scale temperature structure of a general circulation model. Simulated altimetric and tomographic data were then taken, and based on these data and on the known model physics we attempted to estimate the statistics of the errors. We found that even with ten years of altimetric data, estimates of errors in the vertical temperature structure remain very uncertain. By way of contrast, simulated tomographic data provide direct measurements of the temperature errors, and hence significant error estimates are possible using only a small number of years of data.
The value of acoustic tomography then is that it can provide time-continuous measurements of temperature along repeated sections. These measurements are particularly suitable for large-scale circulation studies because they suppress the mesoscale variability which cannot be resolved in global-scale optimizations. I will illustrate the power of acoustic tomography for large scale circulation studies using results from two recent experiments : THETIS-2 in the Western Mediterranean and ATOC in the North Pacific.
