A Multivariate Reduced-order Optimal Interpolation Method and its Application
301
configuration changes in the assimilation. It was found very difficult to test the
ideas put forward in section 15.2 in such a configuration.
Moving to the higher-resolution model, even if still only eddy-permitting given
the small scales of the Rossby radii in the Mediterranean, changed everything as far
as the "response" of the assimilation system was concerned. In practice, the 118°,
3l-level model was found to "keep" the information inserted by the assimilation.
An example is given in Fig. 15.8, showing the Crete-Rhodes area with both resolutions. The small-scale but energetic Ierapetra gyre shows up at the at the southeastern tip of Crete (26°E, 34.5°N) at the end of summer almost every year. It is not
present in the simulations without assimilation at any of the resolutions tested,
because its origin is thought to be related to unresolved small-scale wind cur!
effects. The Ierapetra gyre is seen by satellite altimetry and corrected by assimilation. However the 114° model can barely resolve it (Fig. l5.8b) and is unable to
evolve it dynamically. Assimilation in the 118° model (Fig. l5.8a) does a much
better job; in particular, it is able to produce a much more realist guess of the gyre
one week after correction.
15.4 A brief summary and perspective
The assimilated results show the general circulation variability ofthe Mediterranean in a way which can be improved, but which is coherent between variables and
fairly realistic in many respects. In particular, recurrent seasonal patterns of variability as well as interannual variations of those patterns not present in the free simulation are evidenced in the '90s.
Tests of sensitivity to the choice of the "simplification operator" S as well as to
more realist guess error correlation are underway, in particular as part ofthe Mediterranean Forecasting System Pilot Project (MFSPP). In MFSPP, altimetry as well
as in situ measurements are being assimilated in real-time for demonstration purposes and to get more scientific insight into the relative benefits of the possible
modus operandi and of the various observations. A higher-resolution (1/20°), truly
eddy-resolving model of the Mediterranean is also being implemented in the
framework of the French MERCATOR project (www.mercator.com.fr/html/information/frame ---'produitJr.html). It is fully coupled with a North Atlantic model of a
similar resolution, and uses the same assimilation approach.
The SOFA Fortran code is available free from the author ( Pierre.DeMey@cnes.fr). An alternate version ofthe assimilation code with isopycnal vertical representation as in Gavart and De Mey (1997) has also been implemented in
the 118° model, where isopycnal vertical EOFs are used as "lifting-lowering"
modes. The SOFA algorithm has very simplified error propagation characteristics.
We have performed ensemble integration tests to show the potential of the method
for retrieving the yet unknown forecast error variances. The method has been
applied in the Western Mediterranean to various situations in winter and spring of
1993, and also in a limited-area coastal model in the Gulf of Lyons and Liguro-
301
configuration changes in the assimilation. It was found very difficult to test the
ideas put forward in section 15.2 in such a configuration.
Moving to the higher-resolution model, even if still only eddy-permitting given
the small scales of the Rossby radii in the Mediterranean, changed everything as far
as the "response" of the assimilation system was concerned. In practice, the 118°,
3l-level model was found to "keep" the information inserted by the assimilation.
An example is given in Fig. 15.8, showing the Crete-Rhodes area with both resolutions. The small-scale but energetic Ierapetra gyre shows up at the at the southeastern tip of Crete (26°E, 34.5°N) at the end of summer almost every year. It is not
present in the simulations without assimilation at any of the resolutions tested,
because its origin is thought to be related to unresolved small-scale wind cur!
effects. The Ierapetra gyre is seen by satellite altimetry and corrected by assimilation. However the 114° model can barely resolve it (Fig. l5.8b) and is unable to
evolve it dynamically. Assimilation in the 118° model (Fig. l5.8a) does a much
better job; in particular, it is able to produce a much more realist guess of the gyre
one week after correction.
15.4 A brief summary and perspective
The assimilated results show the general circulation variability ofthe Mediterranean in a way which can be improved, but which is coherent between variables and
fairly realistic in many respects. In particular, recurrent seasonal patterns of variability as well as interannual variations of those patterns not present in the free simulation are evidenced in the '90s.
Tests of sensitivity to the choice of the "simplification operator" S as well as to
more realist guess error correlation are underway, in particular as part ofthe Mediterranean Forecasting System Pilot Project (MFSPP). In MFSPP, altimetry as well
as in situ measurements are being assimilated in real-time for demonstration purposes and to get more scientific insight into the relative benefits of the possible
modus operandi and of the various observations. A higher-resolution (1/20°), truly
eddy-resolving model of the Mediterranean is also being implemented in the
framework of the French MERCATOR project (www.mercator.com.fr/html/information/frame ---'produitJr.html). It is fully coupled with a North Atlantic model of a
similar resolution, and uses the same assimilation approach.
The SOFA Fortran code is available free from the author ( Pierre.DeMey@cnes.fr). An alternate version ofthe assimilation code with isopycnal vertical representation as in Gavart and De Mey (1997) has also been implemented in
the 118° model, where isopycnal vertical EOFs are used as "lifting-lowering"
modes. The SOFA algorithm has very simplified error propagation characteristics.
We have performed ensemble integration tests to show the potential of the method
for retrieving the yet unknown forecast error variances. The method has been
applied in the Western Mediterranean to various situations in winter and spring of
1993, and also in a limited-area coastal model in the Gulf of Lyons and Liguro-