Observing-System Research and Ensemble Data Assimilation at JAMSTEC - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 505

Fig. 21.2 The data flow chart of ALEDAS2. Rectangles represent data, and round rectangles
represent processes

.r/ D exp

2

r

1
2

w (21.7)

where is the localization length parameter. In this modified algorithm, the analysis

is conducted at each model grid point. The reduced matrix size contributes to two to
threefold gains in speed ( Miyoshi et al. 2007b ).

21.2.3 The Forecast-Analysis Cycle

Figure 21.2 shows the data flow chart of ALEDAS2. In the forecast step, each
ensemble member is integrated in time with AFES from the initial conditions input
from an IC file to produce a restart file. AFES is capable of running multiple
ensemble members with a single MPI execution. The restart contains hourly
forecasts at 3-9 h (˙ 3 h from analysis time

) from the initial time. Each restart is
split into seven files for input into LETKF. In the analysis step, observations ( obs )
are assimilated into the forecast in a 6-h window by LETKF to produce analysis
( analysis ). Analysis is performed locally in parallel with MPI processes. The
guess files from LETKF represent the forecast at analysis time

t

in the restart
files. Finally, the analysis files from LETKF replace the model forecast in the
part of the restart files for analysis time

t

t to produce the next initial condition.

21.2.4 Configurations of ALERA2

ALEDAS2 is used to produce the control run for ALERA2 and OSEs. Table 21.2
summarizes the configurations of the two systems. AFES in ALEDAS2 uses
a slightly coarser horizontal resolution and the same vertical resolution, but it
produces a better forecast because of the improved physics described in Sect. 21.2.1 .