Geoscience Reference
In-Depth Information
37.4
37.2
37
36.8
36.6
36.4
36.2
36
35.8
35.6
37.4
37.2
37
36.8
36.6
36.4
36.2
36
35.8
35.6
37.4
37.2
37
36.8
36.6
36.4
36.2
36
35.8
35.6
Ano Nuevo
Ano Nuevo
Ano Nuevo
Pt.Sur
Pt.Sur
Pt.Sur
−
123.2
−
123
−
122.8
−
122.6
−
122.4
−
122.2
−
122
−
121.8
−
121.6
−
121.4
−
123.2
−
123
−
122.8
−
122.6
−
122.4
−
122.2
−
122
−
121.8
−
121.6
−
121.4
−
123.2
−
123
−
122.8
−
122.6
−
122.4
−
122.2
−
122
−
121.8
−
121.6
−
121.4
Longitude (deg W)
Longitude (deg W)
Longitude (deg W)
Fig. 15.6
All glider and the two mooring positions (
left
), and assimilated Slocum (
center
)and
Spray (
right
) glider tracks during August, 2003. The
red dots
represent the location of the moored
buoys M1 (
right
)andM2(
left
). The
red box
indicates the upwelling center near Ano Nuevo
15.4.4
Results with AOSN II Data
The assimilation covers the time window of August 2 to August 27, 2003, and is
carried out in cycles of 5 days, with the analysis at the end of a cycle becoming the
initial condition for the following cycle. Although the observations are processed
and stored in 6-h intervals, the 4D-Var system assimilates all observations within
the 5-day cycle simultaneously. The performance of the assimilation system is
examined by computing the difference between the observations and three model
solutions: (1) the free running (non assimilative) model that is integrated from the
given initial conditions and forcing fields, (2) the first guess (also non assimilative)
for which the initial condition is updated from the assimilation in the previous
cycle, with the exception of the first cycle where both the first guess and the free
running model are equal, and (3) the analysis. The first guess is also the background
trajectory for the tangent linear model and the adjoint, i.e. the trajectory around
which the model is linearized. It is stored in intervals of 6 h. It is anticipated that
due to the re-initialization from assimilating in a previous cycle, the first guess
should have smaller discrepancies with the observations than the free running
model, and the analysis should have smaller discrepancies with the observations
than the first guess. This should be the case for discrepancies computed with the
assimilated and non-assimilated observations. It is expected of every assimilation
system to fit the assimilated observations within one observation standard deviation.
Unassimilated observations consist of withheld observations within the current
assimilation window and future observations, those in the next cycle before the
assimilation. The assimilation is expected to fit the former as a measure of the
system's ability to propagate the information from the assimilated observations sites
through the model spate-time domain within the assimilation window. However,
there is no expectation to fit future observations, i.e. the innovations in the next
cycle are not expected to be smaller than the observation standard deviation. One
only hopes that having initialized the model from the previous cycle's assimilation,
the model forecast will remain sufficiently accurate to maintain small innovations.
However, integrating the model from the initial conditions with uncorrected forcing
fields is prone to drive the model away from the observations.
Search WWH ::
Custom Search