Estimation of oceanographic profiles and climatological regions in the Barents Sea

Vertical profiles of temperature and salinity, or sound speed are used in numerous applications where accurate vertical profiles are crucial. Conventional climatological representations of vertical oceanographic profiles are based on mean or median profiles of historic data in a rectangular area containing the position in question. In areas containing oceanographic fronts mean profiles may not be representative for the profiles in the area and may even be unphysical. Hjelmervik and Hjelmervik [1] proposed a different approach to generate more realistic climatological estimates of the vertical profiles at a given time and position. The depth-dependent behaviours of all historic temperature and salinity profiles are classified by combining Empirical Orthogonal Function (EOF) analysis with K-means clustering. All profiles with similar EOF-coefficients are sorted into a single cluster and averaged to find a representative profile for that cluster. The geographical extent and temporal validity of the cluster are given by the positions and measurement times of the contained profiles. A noticeable disadvantage with this approach is that the EOF approach requires the same number of depth steps in all profiles used. In an area with varying topography, different depth resolutions must be used for different positions. Originally only profiles extending down to a minimum depth were taken into account, and all deeper measurements were discarded. Here we propose using the full profile by employing terrain-following coordinates. Results using terrain-following coordinates are compared to results obtained using a simple approach where each profile is extrapolated in order to obtain a uniform depth resolution. It is shown that the choice of depth representation has a lower impact on the accuracy of the method than for instance the choice of weights applied on the input data, e. g. the relative weights of salinity, temperature, and position input.