374
DAVID ANDERSON ET AL
not the time of greatest wind error. Data assimilation acts to reduce these
differences (not shown). Further illustration of the extent to which data
4.
Weaknesses in the ocean assimilation strategy
The hypothesis underlying the assimilation strategy is that the system is
unbiased. This hypothesis is definitely not true in the case of ocean data
assimilation. This can be seen by evaluating the mean increment applied in
the assimilation. Fig 6 shows this for a section along the equator in the
upper 400m. The upper panel shows the mean increment in temperature,
averaged over eight years. Far from this being zero, one can see that the
assimilation acts to warm the ocean in the west and to cool it in the east
Pacific i.e. to strengthen the gradient along the equator. There is a strong
systematic effect in the Atlantic too but not much in the Indian ocean (which
might just reflect the fact that there are few observations in the Indian
ocean).
If one makes the velocity correction mentioned earlier, this bias is
reduced somewhat. Including a salinity correction also acts to reduce the
bias. Nonetheless, regardless of these changes the bias remains substantial.
It could result from error in the wind, in the ocean model physics, in the way
that momentum is transferred from atmosphere to ocean, or in the
assimilation system itself. Whatever the reason, the assimilation system will
operate at reduced efficiency since it is not designed to deal with bias. In
fact most assimilation schemes assume the system is unbiased. Can we
somehow adapt the assimilation system to take account of the bias? This has
been considered by Dee and Da Silva 1998, Dee and Todling 2000, Vidard
et al 2004, and Bell et al 2004. One approach is to correct the pressure
gradient as suggested by Bell et al 2004. One might think that it would be
better to correct the temperature since that is the field which seems to be
biased, but this is not the case. The vertical velocity along the equator is
distorted (not shown). It is not possible to show what this field should really
look like since it is hard to measure, but a descending circulation in the east
Pacific is generated which looks very unlikely. It is thought that this arises
from the assimilation cycle itself. (See also Vialard et al 2003, Huddleston
et al 2004). Correcting bias in the pressure gradient greatly reduces this
spurious circulation and the mean temperature increment (not shown). On
the other hand correcting the bias in T actually aggravates the situation and
leads to an enhanced spurious vertical circulation, although the T increment
is reduced.
assimilation can act to reduce the impact of wind error is shown in Vialard
et al 2005.
DAVID ANDERSON ET AL
not the time of greatest wind error. Data assimilation acts to reduce these
differences (not shown). Further illustration of the extent to which data
4.
Weaknesses in the ocean assimilation strategy
The hypothesis underlying the assimilation strategy is that the system is
unbiased. This hypothesis is definitely not true in the case of ocean data
assimilation. This can be seen by evaluating the mean increment applied in
the assimilation. Fig 6 shows this for a section along the equator in the
upper 400m. The upper panel shows the mean increment in temperature,
averaged over eight years. Far from this being zero, one can see that the
assimilation acts to warm the ocean in the west and to cool it in the east
Pacific i.e. to strengthen the gradient along the equator. There is a strong
systematic effect in the Atlantic too but not much in the Indian ocean (which
might just reflect the fact that there are few observations in the Indian
ocean).
If one makes the velocity correction mentioned earlier, this bias is
reduced somewhat. Including a salinity correction also acts to reduce the
bias. Nonetheless, regardless of these changes the bias remains substantial.
It could result from error in the wind, in the ocean model physics, in the way
that momentum is transferred from atmosphere to ocean, or in the
assimilation system itself. Whatever the reason, the assimilation system will
operate at reduced efficiency since it is not designed to deal with bias. In
fact most assimilation schemes assume the system is unbiased. Can we
somehow adapt the assimilation system to take account of the bias? This has
been considered by Dee and Da Silva 1998, Dee and Todling 2000, Vidard
et al 2004, and Bell et al 2004. One approach is to correct the pressure
gradient as suggested by Bell et al 2004. One might think that it would be
better to correct the temperature since that is the field which seems to be
biased, but this is not the case. The vertical velocity along the equator is
distorted (not shown). It is not possible to show what this field should really
look like since it is hard to measure, but a descending circulation in the east
Pacific is generated which looks very unlikely. It is thought that this arises
from the assimilation cycle itself. (See also Vialard et al 2003, Huddleston
et al 2004). Correcting bias in the pressure gradient greatly reduces this
spurious circulation and the mean temperature increment (not shown). On
the other hand correcting the bias in T actually aggravates the situation and
leads to an enhanced spurious vertical circulation, although the T increment
is reduced.
assimilation can act to reduce the impact of wind error is shown in Vialard
et al 2005.