Abstract
A method for extending upper ocean density observations to the deep ocean is tested using a large number of deep CTD (conductivity-temperature-depth) stations in the California Current. The specific problem considered is that of constructing the best estimate for the density profile below a certain depth D given an observed profile above that depth. For this purpose, the estimated disturbance profile is modeled as a weighted sum of empirical vertical modes (E0Fs). The EOFs are computed from the surface to 2000 m, using 126 largely independent CTP stations off Point Sur, California. Separate computations are made for the summer half-year (mid-April to mid-October) and the winter half-year (mid-October to mid-April). For each observed density profile. the EOF weights that determine the estimated profile are obtained by performing a successive least-squares fit of the disturbance density profile above D to the first N EOFs. In this study, N is taken to be 7, which is the number of EOFs that account for the “signal” in the profiles as determined by the methods of Preisendorfer et al. and Smith et al. The estimated profiles are then verified against the observed profiles to 2000 m, and the results are presented as a function of the depth D.
In general, the vertical extension method is moderately successful at estimating density fluctuations at and below 500 m from data entirely above 500 m. Observed density profiles to depths shallower than 500 m can he extended to 500 m, with a correlation that depends on the time of year as well as on the depth of the observed profile. For example, a minimum of 200 m of data is needed to perform a useful extension to 500 m, and in all cases extensions are more successful in winter than in summer. As might be expected, correlations between the estimated profiles and a seven-mode reconstruction of the observed profiles, representing the “signal” part of the observed profiles, are somewhat higher. The dynamic height of the sea surface relative to 500 m, an important integral quantity, can be estimated quite well with only 300 m of data. A practical result of this study is that data down to only 200 or 300 m, as might be acquired by a SeaSoar CTD survey, can be extended to 500 m or more using the EOF-based method with a known and useful level of skill. Tests with a small sample of independent data confirm the above results. The success of the method is attributed to the fact that in this part of the ocean the dominant EOFs represent variability in the upper ocean that is also reflected at deeper depths.
