than nominal value as the beam passes through the
scattering layer. The Doppler shift due to the forward motion of the ship is therefore first biased
low, then high, resulting in a characteristic ‘S’ in
the forward velocity component. The height of the
‘S’ is typically a few tens of metres, and the amplitude may be 10–20 cm s
91 (E. Firing, unpublished
data). Localized biases due to coherent swimming
of scattering layers have also been identified near
topography (Wilson and Firing, 1992) and in the
open ocean.
It is perhaps remarkable that the main sources
of SADCP error are almost all due to factors other
than the sonar instrument itself. The exceptions
are the skew errors caused by the tracking algorithm in an early version of the firmware (Chereskin and Harding, 1993), and the scattering layer
‘S’, which is a function of beam width.
3.1.3.3 Lowered ADCP systems
Although the LADCP and SADCP systems and
algorithms are similar in many ways, there are
some important differences relevant to LADCP
profile errors; the LADCP depends much more critically on the inherent accuracy of the sonar. It is
important to distinguish between the accuracy of
the relative velocity profile as a function of vertical
wavenumber, and the accuracy of the depthaveraged velocity – their error sources and characteristics are nearly independent. One must also
distinguish instrumental errors, caused by the fundamental limitations of the hardware and software,
from errors or shortcomings associated with the
way the LADCP samples the ocean.
To avoid confusion in the subsequent discussion, we will use the word ‘profile’ to refer to the
composite profile measured over the whole water
column. Each segment of profile stored by the
instrument will be referred to as an ‘ensemble’.
Generally, an ensemble will result from just one or
two pings.
Because the relative velocity profile is calculated
as the depth-integral of a composite shear profile,
relative velocity errors between two depths tend to
grow as the square root of the separation, as in a
random walk (Fischer and Visbeck, 1993). The
velocity error wavenumber spectrum is red for
scales larger than the depth range of each individual ADCP ensemble, and white for smaller scales
(Firing and Gordon, 1990). The magnitude of the
error increases with uncertainty in the raw ADCP
velocity estimates, and decreases with increasing
range of the individual ensembles and with
increasing numbers of ensembles. This analysis
assumes unbiased ADCP ensembles. Unfortunately, the composite relative velocity profile is
extremely sensitive to small shear biases in the
individual ensembles; fortunately, such bias has
been dominant only in a small fraction of the profiles that have been made. The reason for these
occasional episodes of bias, which cause the downcast and upcast profiles to cross in a characteristic
‘X’ on plots of velocity versus depth, has not yet
been determined. Sometimes the problem can be
reduced by rejecting data from the most distant
bins in each of the single-ping ensembles; sometimes it can be eliminated by rejecting shears from
the top few bins of each ping on the upcast where
a beam intersects the wake of the package. Sometimes none of these is effective and the poor profile
remains unexplained.
Because the relative velocity profile accuracy
depends on the accuracy and the range of the
single-ping ADCP ensembles, it decreases with
reduced acoustic backscattering strength. Backscattering at the 150–300 kHz frequencies typical of
LADCPs varies widely with depth and location. It
generally decreases from the upper ocean to the
abyss, often with a sharp change near 1000 m.
Typical differences exceed 20 db. At all depths,
scattering tends to be weak in the tropics and subtropics, increasing slightly at the equator (particularly in the eastern Pacific) and increasing greatly
in subpolar regions. Consequently, it tends to be
easiest to get good LADCP profiles at high latitudes; conversely, in some low-latitude regions,
the relative velocity profiles have been rendered
useless below about 1000 m. (An estimate of the
global distribution of backscattering strength is
an interesting side effect of WOCE LADCP measurements that is being pursued, but we will not
discuss it further here.)
Until recently, relative velocity profiles were
subject to major interference from sound reflected
from the ocean bottom. For each individual ping
in the affected depth range, the bottom reflection
of the previous ping overwhelms the signal scattered from the water. With a 1-Hz ping rate, for
example, the interference would be centred at
about 650 m off the bottom, and could contaminate a depth band up to 200 m thick. The velocity
signature of this interference depends on the
3.1 Shipboard Observations during WOCE
115
King, Firing and Joyce
scattering layer. The Doppler shift due to the forward motion of the ship is therefore first biased
low, then high, resulting in a characteristic ‘S’ in
the forward velocity component. The height of the
‘S’ is typically a few tens of metres, and the amplitude may be 10–20 cm s
91 (E. Firing, unpublished
data). Localized biases due to coherent swimming
of scattering layers have also been identified near
topography (Wilson and Firing, 1992) and in the
open ocean.
It is perhaps remarkable that the main sources
of SADCP error are almost all due to factors other
than the sonar instrument itself. The exceptions
are the skew errors caused by the tracking algorithm in an early version of the firmware (Chereskin and Harding, 1993), and the scattering layer
‘S’, which is a function of beam width.
3.1.3.3 Lowered ADCP systems
Although the LADCP and SADCP systems and
algorithms are similar in many ways, there are
some important differences relevant to LADCP
profile errors; the LADCP depends much more critically on the inherent accuracy of the sonar. It is
important to distinguish between the accuracy of
the relative velocity profile as a function of vertical
wavenumber, and the accuracy of the depthaveraged velocity – their error sources and characteristics are nearly independent. One must also
distinguish instrumental errors, caused by the fundamental limitations of the hardware and software,
from errors or shortcomings associated with the
way the LADCP samples the ocean.
To avoid confusion in the subsequent discussion, we will use the word ‘profile’ to refer to the
composite profile measured over the whole water
column. Each segment of profile stored by the
instrument will be referred to as an ‘ensemble’.
Generally, an ensemble will result from just one or
two pings.
Because the relative velocity profile is calculated
as the depth-integral of a composite shear profile,
relative velocity errors between two depths tend to
grow as the square root of the separation, as in a
random walk (Fischer and Visbeck, 1993). The
velocity error wavenumber spectrum is red for
scales larger than the depth range of each individual ADCP ensemble, and white for smaller scales
(Firing and Gordon, 1990). The magnitude of the
error increases with uncertainty in the raw ADCP
velocity estimates, and decreases with increasing
range of the individual ensembles and with
increasing numbers of ensembles. This analysis
assumes unbiased ADCP ensembles. Unfortunately, the composite relative velocity profile is
extremely sensitive to small shear biases in the
individual ensembles; fortunately, such bias has
been dominant only in a small fraction of the profiles that have been made. The reason for these
occasional episodes of bias, which cause the downcast and upcast profiles to cross in a characteristic
‘X’ on plots of velocity versus depth, has not yet
been determined. Sometimes the problem can be
reduced by rejecting data from the most distant
bins in each of the single-ping ensembles; sometimes it can be eliminated by rejecting shears from
the top few bins of each ping on the upcast where
a beam intersects the wake of the package. Sometimes none of these is effective and the poor profile
remains unexplained.
Because the relative velocity profile accuracy
depends on the accuracy and the range of the
single-ping ADCP ensembles, it decreases with
reduced acoustic backscattering strength. Backscattering at the 150–300 kHz frequencies typical of
LADCPs varies widely with depth and location. It
generally decreases from the upper ocean to the
abyss, often with a sharp change near 1000 m.
Typical differences exceed 20 db. At all depths,
scattering tends to be weak in the tropics and subtropics, increasing slightly at the equator (particularly in the eastern Pacific) and increasing greatly
in subpolar regions. Consequently, it tends to be
easiest to get good LADCP profiles at high latitudes; conversely, in some low-latitude regions,
the relative velocity profiles have been rendered
useless below about 1000 m. (An estimate of the
global distribution of backscattering strength is
an interesting side effect of WOCE LADCP measurements that is being pursued, but we will not
discuss it further here.)
Until recently, relative velocity profiles were
subject to major interference from sound reflected
from the ocean bottom. For each individual ping
in the affected depth range, the bottom reflection
of the previous ping overwhelms the signal scattered from the water. With a 1-Hz ping rate, for
example, the interference would be centred at
about 650 m off the bottom, and could contaminate a depth band up to 200 m thick. The velocity
signature of this interference depends on the
3.1 Shipboard Observations during WOCE
115
King, Firing and Joyce