ICHIRO FUKUMORI
330
heoretical aspects of these and other approximations. Here
we
ng process noise
ferent model errors sources could be
considered independent of one another. Then different process noise and
thei
s in large-scale wind
can
e errors. This estimate should not
be c
s in temporal and spatial scales, wind-driven
barotropic errors could be considered independent of baroclinic errors and
thus
problem’s degrees of freedom without incurring excessive computational
requirements.
The reader is referred to Chapter 11 and to references above for further
discussion on t
review examples of implementing the approximations and their
implications.
5.2.1 Identifyi
To first approximation, dif
r consequent model state errors could be evaluated separately in the
context of a partitioned estimation (Fukumori, 2002).
Different sources of process noise cause different errors in the modeled
state. For instance, the response of a model to change
be effectively described in terms of the gravest few vertical dynamic
modes (e.g., Cane, 1984). In comparison, a model’s response to changes in
air-sea heat flux is to first approximation confined to the sea surface. The
modeled process noise dictates the most effective state approximation (e.g.,
state reduction and partitioning), and, therefore, its identification is the first
step in designing an assimilation system.
The ongoing ECCO near real-time assimilation estimates uncertainties
of wind forcing and its resulting model stat
onfused as one that considers all model errors are due to errors in wind,
but it is an estimate of only a part of the errors, as discussed above, albeit
one of the dominant ones. The model’s controllability (ability to uniquely
solve u in Eq 1) limits aliasing of other model error sources to the particular
process noise being estimated. The ECCO near real-time assimilation
system described below is correspondingly designed to resolve the dominant
response of the ocean to large-scale wind errors.
5.2.2 Regional partitions
Due to large difference
estimated separately. Having sub-basin length scales, the baroclinic
components are estimated individually among seven different basins across
the globe (Figure 3). These regions include three separate tropical basins
(Indian, Pacific, Atlantic) and four mid- and high-latitude basins (North
Pacific, North Atlantic, South Atlantic and Indian, South Pacific). The
regions overlap each other to minimize edge effects caused by the regional
approximation; errors in overlapping areas are considered to be split among
330
heoretical aspects of these and other approximations. Here
we
ng process noise
ferent model errors sources could be
considered independent of one another. Then different process noise and
thei
s in large-scale wind
can
e errors. This estimate should not
be c
s in temporal and spatial scales, wind-driven
barotropic errors could be considered independent of baroclinic errors and
thus
problem’s degrees of freedom without incurring excessive computational
requirements.
The reader is referred to Chapter 11 and to references above for further
discussion on t
review examples of implementing the approximations and their
implications.
5.2.1 Identifyi
To first approximation, dif
r consequent model state errors could be evaluated separately in the
context of a partitioned estimation (Fukumori, 2002).
Different sources of process noise cause different errors in the modeled
state. For instance, the response of a model to change
be effectively described in terms of the gravest few vertical dynamic
modes (e.g., Cane, 1984). In comparison, a model’s response to changes in
air-sea heat flux is to first approximation confined to the sea surface. The
modeled process noise dictates the most effective state approximation (e.g.,
state reduction and partitioning), and, therefore, its identification is the first
step in designing an assimilation system.
The ongoing ECCO near real-time assimilation estimates uncertainties
of wind forcing and its resulting model stat
onfused as one that considers all model errors are due to errors in wind,
but it is an estimate of only a part of the errors, as discussed above, albeit
one of the dominant ones. The model’s controllability (ability to uniquely
solve u in Eq 1) limits aliasing of other model error sources to the particular
process noise being estimated. The ECCO near real-time assimilation
system described below is correspondingly designed to resolve the dominant
response of the ocean to large-scale wind errors.
5.2.2 Regional partitions
Due to large difference
estimated separately. Having sub-basin length scales, the baroclinic
components are estimated individually among seven different basins across
the globe (Figure 3). These regions include three separate tropical basins
(Indian, Pacific, Atlantic) and four mid- and high-latitude basins (North
Pacific, North Atlantic, South Atlantic and Indian, South Pacific). The
regions overlap each other to minimize edge effects caused by the regional
approximation; errors in overlapping areas are considered to be split among