THE ECMWF PERSPECTIVE
375
Figure 6. Assimilation increment averaged over an 8-year period in a vertical section along
the equator. The left panel is the Indian ocean, the middle the Pacific and the right is the
Atlantic. In a well-balanced system the mean increment should be close to zero. In practice
most ocean assimilation systems have considerable bias, as shown here. The assimilation is
acting to warm the west Pacific but to cool the east. Some of the bias is caused by the
assimilation system itself. Warming regions, i.e. where the assimilation increment is positive,
are shaded. Contour interval is 2K/year.
5.
Ocean observing system experiments
As funding is always limited, the question of the relative merit of
each observational system arises. This can be estimated through observation
system experiments (OSEs), well known to meteorologists and described in
fig 1. See Daley 1992 for general discussion and Anderson et al 1991 and
Isaksen and Stoffelen 2000 for discussion relevant to surface wind field from
scatterometers. In oceanography, this is a relatively new field, as
observations have always been sparse.
The fairest way to evaluate the potential of an observing system is to
selectively withdraw components of the system starting with the full array.
So we take the system with TAO/PIRATA/TRITON moorings, the XBT
network and the ARGO array as the standard, and then remove,
TAO/PIRATA/TRITON. The impact of this component of the observing
system can then be assessed. The XBTs can be assessed by removing them
and comparing with the standard system. This strategy is different to
starting with a zero system and then adding components. The approach used
here allows for redundancy. So, it might be that a component of the array
can be withdrawn at little overall effect because more or less the same
information is available from another component of the observing system.
Some redundancy is not necessarily bad, however, as it allows an evaluation
of the different parts of the array. It should also be remembered that even
though there is some redundancy in some average sense, there might be
375
Figure 6. Assimilation increment averaged over an 8-year period in a vertical section along
the equator. The left panel is the Indian ocean, the middle the Pacific and the right is the
Atlantic. In a well-balanced system the mean increment should be close to zero. In practice
most ocean assimilation systems have considerable bias, as shown here. The assimilation is
acting to warm the west Pacific but to cool the east. Some of the bias is caused by the
assimilation system itself. Warming regions, i.e. where the assimilation increment is positive,
are shaded. Contour interval is 2K/year.
5.
Ocean observing system experiments
As funding is always limited, the question of the relative merit of
each observational system arises. This can be estimated through observation
system experiments (OSEs), well known to meteorologists and described in
fig 1. See Daley 1992 for general discussion and Anderson et al 1991 and
Isaksen and Stoffelen 2000 for discussion relevant to surface wind field from
scatterometers. In oceanography, this is a relatively new field, as
observations have always been sparse.
The fairest way to evaluate the potential of an observing system is to
selectively withdraw components of the system starting with the full array.
So we take the system with TAO/PIRATA/TRITON moorings, the XBT
network and the ARGO array as the standard, and then remove,
TAO/PIRATA/TRITON. The impact of this component of the observing
system can then be assessed. The XBTs can be assessed by removing them
and comparing with the standard system. This strategy is different to
starting with a zero system and then adding components. The approach used
here allows for redundancy. So, it might be that a component of the array
can be withdrawn at little overall effect because more or less the same
information is available from another component of the observing system.
Some redundancy is not necessarily bad, however, as it allows an evaluation
of the different parts of the array. It should also be remembered that even
though there is some redundancy in some average sense, there might be