### "Salting" South Atlantic XBTs

The regression models computed within overlapping shingles provide
estimates of salinity to accompany XBT profiles within non-overlapping
4x1 regions at regularly spaced depths. A choice of 16 different
models for each target region and each level is available. When
compared against independent verification data the models using
temperature, temperature squared, longitude, and latitude were among
the best for all regions and for all depths.

The regression models can be represented as a table of coefficients
and indicators of the region and level at which the coefficients
apply. For the model employing all four regressors, salinity can be
computed as:

S = a + b × T + c × T^{2} + d × longitude + e × latitude

Where **longitude** and **latitude** are the coordinates of the XBT station, **T**
is the temperature at one of the standard pressure levels at 25 dbar
intervals, **a** through **e** are the regression coefficients appropriate for
that level for the shingle with center closest to the XBT station, and
where **S** is the estimate for salinity at that station and level. The
model coefficients can be tabulated along with information about the
pressure level and the longitude and latitude of the center of the
shingle to which they apply.

XBT temperatures are usually reported for depths determined from the
fall time of the probe, which can be converted to pressure using the
hydrostatic equation with the climatological density profile in the
vicinity of the station. Once salinity values have been estimated
from the measured temperatures, an improved density profile can be
computed from the equation of state of sea-water to get an improved
estimate of pressure to be used for a corrected estimate of salinity.

To get estimates at other levels, the regression coefficients can be
interpolated from the standard levels to the desired levels, e.g. the
levels where the XBT temperatures were recorded. Because surface
processes make surface salinity very difficult to estimate, no
regression coefficients were given for 0 dbar. Near-surface estimates
might be made using the models for 25 dbar.