NESTING OCEAN MODELS
143
algorithm in some very few iterations. Development of such algorithms
for realistic ocean models is ongoing research work.
5.
Software tools
Designing nested or coupled systems starting from existing models
is quite a difficult and time-consuming practical task. However several
software tools have been developed these last years, which automatically
manage an important part of the job.
The AGRIF package1 (Debreu et al., 2004a) allows an easy integration of mesh refinement capabilities within any existing finitedifference
model written in Fortran. One can therefore design one-way and twoway multiply-nested systems, with the possibility of adaptive regridding,
without reprogramming the model. This package is presently implemented into several operational ocean models.
General couplers can also be used to implement nested systems, especially in the case when the local and external model codes are totally
different. The user has then to prescribe the structure of the coupling
algorithm and the interactions between the different objects, but at a
rather high level, without having to go too much into programming
details. In the context of geophysical fluids, we can cite for instance
PALM^ or M ~ C C I ~ .
6.
Data assimilation and nesting
Along with the development of nested ocean modelling systems, the
problem of assimilating data within these systems is presently strongly
emerging. Addressing this difficult problem is out of the scope of this
note. Let us however point out a few related issues.
The exact mathematical formulation of the data assimilation problem for one-way or tweway nested systems is far from obvious. A
first attempt in this direction for the 4D-Var approach can be
found in Debreu et al. (2004b). Concerning the stochastic approach, interesting ideas can probably be found in the theories of
multiresolution stochastic models and multiscale estimation.
Several ad-hoc procedures are already in use in numerous systems.
A possibility is to perfom the assimilation only on one grid (the
largest or the finest) of the system. Another way is to "hide" the
lhttp: / / m - l m c . imag. f r/IDOPT/AGRIF
2http://~.cerfacs .fr/-palm
3http://m.mpcci .org
143
algorithm in some very few iterations. Development of such algorithms
for realistic ocean models is ongoing research work.
5.
Software tools
Designing nested or coupled systems starting from existing models
is quite a difficult and time-consuming practical task. However several
software tools have been developed these last years, which automatically
manage an important part of the job.
The AGRIF package1 (Debreu et al., 2004a) allows an easy integration of mesh refinement capabilities within any existing finitedifference
model written in Fortran. One can therefore design one-way and twoway multiply-nested systems, with the possibility of adaptive regridding,
without reprogramming the model. This package is presently implemented into several operational ocean models.
General couplers can also be used to implement nested systems, especially in the case when the local and external model codes are totally
different. The user has then to prescribe the structure of the coupling
algorithm and the interactions between the different objects, but at a
rather high level, without having to go too much into programming
details. In the context of geophysical fluids, we can cite for instance
PALM^ or M ~ C C I ~ .
6.
Data assimilation and nesting
Along with the development of nested ocean modelling systems, the
problem of assimilating data within these systems is presently strongly
emerging. Addressing this difficult problem is out of the scope of this
note. Let us however point out a few related issues.
The exact mathematical formulation of the data assimilation problem for one-way or tweway nested systems is far from obvious. A
first attempt in this direction for the 4D-Var approach can be
found in Debreu et al. (2004b). Concerning the stochastic approach, interesting ideas can probably be found in the theories of
multiresolution stochastic models and multiscale estimation.
Several ad-hoc procedures are already in use in numerous systems.
A possibility is to perfom the assimilation only on one grid (the
largest or the finest) of the system. Another way is to "hide" the
lhttp: / / m - l m c . imag. f r/IDOPT/AGRIF
2http://~.cerfacs .fr/-palm
3http://m.mpcci .org