Delayed-mode calibration of profiling float salinity data by historical hydrographic data

The ARGO program calls for a deployment of 3000 profiling floats throughout the world ocean. These floats will drift at intermediate depth, providing estimates of the flow field there. Every 10 days the floats will rise to the surface to report their positions and observed oceanic conditions in the upper 2000 m of the water column. Each float will make a vertical profile of temperature and salinity as a function of pressure as the float rises to the surface. The temperature and pressure sensors are expected to be relatively stable and accurate over the 4-year lifetime of the floats. However, the salinity measurements may accumulate significant errors as the sensors drift away from their initial calibrations for a variety of reasons. Logistics preclude retrieval of these floats for laboratory calibration to monitor sensor drift. Therefore, a system is under development to compare and calibrate the salinity data from the profiling floats with nearby historical data.

A group of nearby historical data with the strongest spatial and temporal correlations to the float profile is selected from the World Ocean Database (1998), providing comparison values of potential temperature (q), salinity (S), and pressure. Correlation time-scales are estimated from a global data set of chlorofluorocarbons (CFCs) that provides apparent age since ventilation (a rough measure of how long ago the water last saw the surface). Spatial-scale estimates are based on the average water mass length scales of the region, and are typically around 8° for longitudinal scale and 4° for latitudinal scale. An objective mapping technique is then used to obtain a q-S climatology (with error estimates) at the location and time of individual float profiles. The parameters describing the conductivity sensor drift, a time-varying slope and an overall bias, are least-squares fitted to the climatology in potential conductivity space, using the errors from the climatological q-S data as the weighting. The result is a set of calibrated salinity data including an uncertainty for the calibration. Because of the need to accumulate a time-series for calculating the calibration coefficients and the desirability of evaluating the fits away from end-points, this system will work best as a delayed-mode quality control system. A delay of a few months is the anticipated time for stable salinity calibrations.