For this study, CTD stations within areal bins in the tropical Pacific are
averaged to make a mean hydrographic profile for each bin. In latitude, the
bins are centered every 1° from 20°S to 20°N. In longitude, the
bins are centered every 5° from 120°E to 75°W. This anisotropy
is dictated by the data distribution, consisting mostly of closely sampled meridional
sections with wide zonal spacing, but is also appropriate to the tropical ocean
interior where zonal scales greatly exceed meridional scales. For this study
15,693 individual CTD stations between 20.5°S and 20.5°N, 120°E
and 70°W and taken from 1967 through 1998 are used. These data are all
the high-resolution profiles available from the National Oceanographic Data
Center (NODC) archives, as well as recent Pacific Marine Environmental Laboratory
(PMEL) cruise data and some other cruise data not yet available from NODC. The
average profiles are used to make maps of properties on neutral density anomaly,
n, surfaces (Jackett
and McDougall 1997), meridional-vertical water property sections, and mass
transport estimates. Only profiles with more than two CTD casts are used for
the property maps and transport estimates and with two or more CTD casts for
the vertical sections.
The CTD station data within each bin, consisting of salinity S and temperature
T as a function of pressure P are averaged as a function of n
to create mean hydrographic profiles. Averaging S, T, and P
as functions of n is more
involved than the conventional approach of averaging S, T, and
n as functions of P,
but this technique better preserves water properties in the sharp tropical pycnocline
and the strength of the pycnocline itself because averaging is quasi-Lagrangian
in the vertical (Gouriou
and Toole 1993). First, to reduce small-scale noise, individual CTD station
S and T profiles are filtered in P with a 5-dbar half-width
Hanning filter and subsampled at 5-dbar intervals. Then n
is computed for these subsampled data and averaged as a function of P
to obtain a mean n(P)
profile at 5-dbar resolution. Following this step, the individually filtered
and sampled profiles of S, T, and P are linearly interpolated
to each n(P) profile
value to allow averaging of the individual CTD station data as a function of
n at roughly 5-dbar resolution.
Mean profiles for S(n),
T(n), and P(n)
are then calculated. Finally the mean P(n)
profile is used to put the mean profiles of S(n)
and T(n) onto an
even 5-dbar grid and construct a final mean n
profile. The profile is limited to values denser than the densest mixed layer
for the CTD stations within the bin.
Depth and properties of the surface mixed layer, including n,
vary over time. To construct a mixed layer for the mean profiles of S(n),
T(n), and n,
the following procedure is used. First, the mixed layer P for each CTD
station is defined as the P above which n
is less than 0.05 kg m
denser than the
mean n of the top 10 dbar.
Mean S and T from the surface to the mixed layer P are
calculated for each CTD station. Then these S and T values are
averaged to find mean mixed layer values for S and T. These values
are used with the mean mixed layer P to calculate a mean mixed layer
n. In order to avoid n
inversions in the mean profiles, the pressure of the mean mixed layer
n in the final mean n
profile is used to define a final mean mixed-layer P. To finish, the
mean mixed layer S, T, and n
values are substituted into the mean S(n),
T(n), and n
profiles from the mean mixed layer P to the surface. The gap between
the mean mixed layer and the mean interior profiles is interpolated using a
shape-preserving local spline (Akima
1970).
The analysis exclusively uses the resulting 5-dbar mean hydrographic profiles.
The only further vertical smoothing performed is on the square of the buoyancy
frequency, N
, which is calculated
from the mean hydrographic profiles and then smoothed with a 35-dbar median
filter before use. The median filter effectively reduces noise while preserving
the rapid transition from the surface mixed layer to the pycnocline. This quantity
is then used to calculate the planetary component of the potential vorticity,
Q = N f / g.
A few Q profiles still have some noise at depth in bins with few stations
and some of those ending in shallow water. Neutral surface maps are made by
linearly interpolating the 5-dbar values of each mean hydrographic profile to
the appropriate n. These
values are then objectively mapped assuming a Gaussian covariance with correlation
length scales of 10° longitude and 2° latitude and a noise-to-signal
variance ratio of 0.25. These correlation length scales are twice the bin size,
to emphasize only those large-scale features that are well resolved by the data.
Return to previous section or go to next section