308
PIERRE BRASSEUR
8.1
Computing innovations using FGAT
Typical lengths of assimilation cycles are 3 to 7 days for mesoscale
ocean current predictions, and 10 to 30 days for initialization of the
oceanic component of seasonal climate predictions. In spite of the fact
that the ocean cannot be considered as static over these time scales,
intermittent sequential filters incorporate at one single instant the set of
observations collected during the whole assimilation interval. This is a
major dierence with 4D-VAR algorithms, which can take full advantage
of the temporal distribution of the data within an assimilation window.
Fairly simple solutions can be set up to alleviate these problems in
the context of statistical filters. For example, the FGAT (First Guess
at Appropriate Time) method initially introduced in meteorology can
be used to evaluate the innovation vector more correctly: instead of
computing the dierence between the time-distributed data set and the
model forecast at time t i+1 as in Eq. (19), the innovation is evaluated
“on the flight” by cumulating the dierences between each piece of observation and the corresponding element of the model forecast at the
measurement time. This approach has also been used with 3D-VAR
assimilation systems [Weaver et al., 2003] to benefit from the temporal
dimension. Due to the fast propagation of equatorial waves, the FGAT
feature may be particularly important in the tropical oceans.
8.2
Incremental Analysis Update
A direct consequence of intermittency is the discontinuity of the forecast/analysis estimates, which is recognized as a major drawback of sequential methods. Two related problems, - shocks to the model and data
rejection -, arise with intermittent corrections. It is found that observations assimilated into models may introduce transient waves excited by
the impulsive insertion. These waves are often the result of imperfections
in the corrected state associated with physically unbalanced error covariances. In the example illustrated in figure 9, six vertical profiles of temperature and salinity measurements are assimilated into the HYCOM
model using the SEEK filter. The profiles are inserted in distant regions
(Labrador Basin, Irminger Sea, Gulf Stream, Azores and North Brazil
currents and Caribbean Sea) in order to avoid mutual interference. The
corrected state is then integrated using the model with realistic forcings
for one month after analysis time. Figure 9 depicts the SSH increment
after 3 days of simulation, showing the occurrence of spurious transients
(particularly in the Gulf Stream region) which the model generates to
dynamically adjust the new state.
PIERRE BRASSEUR
8.1
Computing innovations using FGAT
Typical lengths of assimilation cycles are 3 to 7 days for mesoscale
ocean current predictions, and 10 to 30 days for initialization of the
oceanic component of seasonal climate predictions. In spite of the fact
that the ocean cannot be considered as static over these time scales,
intermittent sequential filters incorporate at one single instant the set of
observations collected during the whole assimilation interval. This is a
major dierence with 4D-VAR algorithms, which can take full advantage
of the temporal distribution of the data within an assimilation window.
Fairly simple solutions can be set up to alleviate these problems in
the context of statistical filters. For example, the FGAT (First Guess
at Appropriate Time) method initially introduced in meteorology can
be used to evaluate the innovation vector more correctly: instead of
computing the dierence between the time-distributed data set and the
model forecast at time t i+1 as in Eq. (19), the innovation is evaluated
“on the flight” by cumulating the dierences between each piece of observation and the corresponding element of the model forecast at the
measurement time. This approach has also been used with 3D-VAR
assimilation systems [Weaver et al., 2003] to benefit from the temporal
dimension. Due to the fast propagation of equatorial waves, the FGAT
feature may be particularly important in the tropical oceans.
8.2
Incremental Analysis Update
A direct consequence of intermittency is the discontinuity of the forecast/analysis estimates, which is recognized as a major drawback of sequential methods. Two related problems, - shocks to the model and data
rejection -, arise with intermittent corrections. It is found that observations assimilated into models may introduce transient waves excited by
the impulsive insertion. These waves are often the result of imperfections
in the corrected state associated with physically unbalanced error covariances. In the example illustrated in figure 9, six vertical profiles of temperature and salinity measurements are assimilated into the HYCOM
model using the SEEK filter. The profiles are inserted in distant regions
(Labrador Basin, Irminger Sea, Gulf Stream, Azores and North Brazil
currents and Caribbean Sea) in order to avoid mutual interference. The
corrected state is then integrated using the model with realistic forcings
for one month after analysis time. Figure 9 depicts the SSH increment
after 3 days of simulation, showing the occurrence of spurious transients
(particularly in the Gulf Stream region) which the model generates to
dynamically adjust the new state.