Contributions to Global Ocean Observations
51
50 temperature-pressure recorders, (b) 20 SOFAR floats, and (c) 12 repeated STD
surveys over a 77-station grid with spacing from 33 to 50 km, all with approximately
uniform coverage over an area roughly 200 km in diameter and augmented by several limited or special-purpose observations. The last MODE effort, POLYMODE,
included long-term time series extending from MODE-1 and substantial geographical exploration using mooring arrays, improved SOFAR floats, and repeated closely
spaced XBT sections.
One of the revolutionary aspects of MODE was the way it was designed. The
summer camp to plan MODE-1 began with vigorous debates about the relative emphasis to be placed on improving the statistical description of the mesoscale (e.g.,
wavenumber–frequency spectra), developing a multivariable synoptic picture that
could be compared with model simulations, and on testing dynamical principles (e.g.,
geostrophy, conservation of potential vorticity). Klaus Hasselman argued for a statistical description, which would drive observation placement toward a uniform co-array
(collection of observation separations). Modelers favored an array that provided
good coverage for initial and boundary conditions with some interior observations
to verify model predictions. Bretherton and Chris Fandry arrived with an appreciation for how Gandin’s (1965) objective analysis could help design sampling arrays
while, coincidentally, I had unwittingly reinvented the same method in Fourier space.
We (Bretherton, Davis, and Fandry, 1976) used objective analysis to address array
measurement accuracy and came to appreciate the way that data-based maps depend
as much on the statistics assumed in mapping as on the data and that the dynamics
these maps imply are set mainly by the statistics. Consequently, dynamical tests are
sometimes best done directly with statistics (see Hogg, 1974).
Eventually common sense held sway over sophisticated arguments. It was decided to accurately map as large an area as possible so that model verification, statistical description, and data-based dynamical testing could all be carried out. Extensive
experimentation with many possible arrays and various statistical representations
based on MODE-0 recommended a roughly uniform array of 200-km diameter. At
the same time, Bill Schmitz at WHOI had designed an array to meet the same general
objective based on a verbal report of the correlation scale and without calculation.
I have always appreciated that the two approaches led to virtually identical arrays;
objective analysis, however, could estimate mapping errors.
The ambitious MODE effort depended on emerging observation capabilities,
but early results showed the limits of those capabilities. MODE-0 used both surface
and intermediate moorings and an important comparison (Gould and Sambuco, 1975)
showed that current meters on surface moorings had significantly elevated kinetic energy, reflecting errors caused by wave-induced mooring motion. It was known that in
oscillatory flow the Savonius rotor accelerated faster than it decelerated, magnifying
speed. New laboratory tests showed that the vane was too slow to completely reverse
in oscillating flow, further magnifying speed, and, most seriously, vertical motion
in the line below surface buoys caused rotors to speed up. After a decade of use,
MODE-0 showed that these errors were so great as to make measured currents little
51
50 temperature-pressure recorders, (b) 20 SOFAR floats, and (c) 12 repeated STD
surveys over a 77-station grid with spacing from 33 to 50 km, all with approximately
uniform coverage over an area roughly 200 km in diameter and augmented by several limited or special-purpose observations. The last MODE effort, POLYMODE,
included long-term time series extending from MODE-1 and substantial geographical exploration using mooring arrays, improved SOFAR floats, and repeated closely
spaced XBT sections.
One of the revolutionary aspects of MODE was the way it was designed. The
summer camp to plan MODE-1 began with vigorous debates about the relative emphasis to be placed on improving the statistical description of the mesoscale (e.g.,
wavenumber–frequency spectra), developing a multivariable synoptic picture that
could be compared with model simulations, and on testing dynamical principles (e.g.,
geostrophy, conservation of potential vorticity). Klaus Hasselman argued for a statistical description, which would drive observation placement toward a uniform co-array
(collection of observation separations). Modelers favored an array that provided
good coverage for initial and boundary conditions with some interior observations
to verify model predictions. Bretherton and Chris Fandry arrived with an appreciation for how Gandin’s (1965) objective analysis could help design sampling arrays
while, coincidentally, I had unwittingly reinvented the same method in Fourier space.
We (Bretherton, Davis, and Fandry, 1976) used objective analysis to address array
measurement accuracy and came to appreciate the way that data-based maps depend
as much on the statistics assumed in mapping as on the data and that the dynamics
these maps imply are set mainly by the statistics. Consequently, dynamical tests are
sometimes best done directly with statistics (see Hogg, 1974).
Eventually common sense held sway over sophisticated arguments. It was decided to accurately map as large an area as possible so that model verification, statistical description, and data-based dynamical testing could all be carried out. Extensive
experimentation with many possible arrays and various statistical representations
based on MODE-0 recommended a roughly uniform array of 200-km diameter. At
the same time, Bill Schmitz at WHOI had designed an array to meet the same general
objective based on a verbal report of the correlation scale and without calculation.
I have always appreciated that the two approaches led to virtually identical arrays;
objective analysis, however, could estimate mapping errors.
The ambitious MODE effort depended on emerging observation capabilities,
but early results showed the limits of those capabilities. MODE-0 used both surface
and intermediate moorings and an important comparison (Gould and Sambuco, 1975)
showed that current meters on surface moorings had significantly elevated kinetic energy, reflecting errors caused by wave-induced mooring motion. It was known that in
oscillatory flow the Savonius rotor accelerated faster than it decelerated, magnifying
speed. New laboratory tests showed that the vane was too slow to completely reverse
in oscillating flow, further magnifying speed, and, most seriously, vertical motion
in the line below surface buoys caused rotors to speed up. After a decade of use,
MODE-0 showed that these errors were so great as to make measured currents little