How useful are localized clusters of traditional oceanographic measurements for data assimilation?

Abstract

Abstract The provocative question we ask in this paper is prompted by the fact that in the excitement induced by the forthcoming availability of the altimetric data set, traditional oceanographic measurements seem to be sometimes underestimated in data assimilation studies. Nevertheless, regional data sets such as those provided by localized clusters of moorings are rich in information, especially for resolving the structure and variability of the deep oceanic layers that are not ‘seen’ by altimetry. Long time series of velocity and temperature measured at different depths over the entire water column in localized clusters of moorings constitute an example of common oceanographic measurements. They are the backbone of important experiments such as SYNOP (Synoptic Ocean Prediction) that focus upon process studies of the Gulf Stream system, the prototype of an energetic western boundary current. Thus we address the question: can these localized clusters of data points be useful and effective when assimilated into an ocean circulation model for process studies in the Gulf Stream system? Can the model internal dynamics, constrained by the limited data sets, reconstruct events such as meander formation and steepening, ring pinch-off and migration in the wide regions that remain void of measurements? To address these questions we have carried out assimilation experiments with a primitive equation model in which the central and eastern arrays of SYNOP are idealized as two identical clusters of 13 moorings with each pair of adjacent moorings approximately 100 km apart. The model configuration is of a zonal channel with inflow/outflow conditions and flat bottom. The twin experiment approach is used. Four types of oceanographic data are considered leading to four different assimilation experiments. They are the time series of: (1) velocity; (2) temperature (density); (3) velocity and density; and (4) total transport. The data are provided at five different levels over the model depth and the assimilation is carried out for 60 days. The success of the assimilation was beyond our expectations. The velocity time series proved to be the most successful type of data. Less than 2 months of assimilated velocity data are sufficient to reconstruct an event of meander formation, steepening, bending and breaking off. The meander event is localized in the region roughly between the two arrays where no data are available. The fact that velocity data are more effective than density ones can be rationalized in terms of transfer of information between different model variables, as knowledge of the density field provides information only of the geostrophic component of the velocity and not of its important ageostrophic part. Thus, these results suggest the need for caution when applying tradional geostrophic adjustment theory to different types of flow regimes. Thus, the answer to the initial question is very positive. Even though the idealized nature of the simulations, and their consequent limitations, must be kept in mind, a cautious optimism seems reasonable for the effectiveness of the real SYNOP data when assimilated in more realistic model configurations.