Abstract
SALARGOS buoys that measure upper-ocean temperature and salinity in ice-covered seas have been collecting data in the Arctic basin for several years. The buoys consist of a 300-m-long string of six temperature-conductivity sensors at fixed depths, with a pressure sensor at the bottom. The electronics housing, including an Argos transmitter, is frozen into the sea ice. The buoy drifts with the sea ice, sampling a region described by the Lagrangian drift of the surrounding ice pack. When the ice is moving slowly, the relative water velocity in the upper ocean is low and the buoy simply produces hydrographic time series at six depths. In periods of rapid drift the buoy cable is displaced upward and the sensors sample other depths. Collecting such data over time can produce plots with relatively high vertical resolution. But to use this information, one needs a reasonable curve fit to these data similar to a smoothed version of a CTD (conductivity-temperature-depth) cast. Two methods have been investigated parametric regression analysis and nonparametric smoothing routines. The quality of the parametric fits depends in part on the choice of analytical profile, as is demonstrated here by comparing the efficacy of a common three-parameter model with a more complete five-parameter model. Both share the advantage of containing explicit geophysical quantities (such as mixed-layer depth) as the free parameters of the system. The nonpammetric smoother, on the other hand, assumes no a priori knowledge of the underlying physics. The data are smoothed and interpolated onto a grid of depth values in a procedure that includes both median and mean running fitters, which results in a relatively small standard error. The standard error for the parametric routines is larger by a factor of 2 or 3, but these schemes produce better estimates of geophysical parameters such as mixed-layer depth. Either technique could also be used to gain enhanced vertical resolution with bottom-moored ocean buoys.
