Development of a Two-way Nested LETKF System for Cloud-resolving Model - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 485

As for the Outer LETKF system, it was used to reproduce mesoscale distributions
including convergence lines. The grid interval of the Outer LETKF is 15 km
and the grid number in the horizontal directions was . The Kain-Fritch

parameterization scheme was adopted. The ensemble forecast started at 0900 JST
1st September 2008 and the initial seed of the Outer LETKF was obtained from the
JMA mesoscale analysis fields from 29th to 31st August. The boundary condition
from 1st to 5th September was also produced from the JMA mesoscale analysis.
The data assimilation window (1 cycle) was 6 h and the conventional data (surface
and upper sounding data), which was used in the JMA mesoscale analysis, were
assimilated every hour.
The Inner LETKF was used to reproduce the intense convection cells of local
heavy rainfalls. The grid interval of the Inner LETKF was as small as 1.875 km
to resolve small convection cells. The microphysical process, in which the mixing
ratio of cloud, rain, ice crystals, graupel and the number density of ice crystals
were predicted, was used. The boundary conditions and first initial seed of the
Inner LETKF (indicated by an open triangle in Fig. 20.4 a) were produced from the
forecast of the Outer LETKF. The data assimilation window (1 cycle) is 1 h, and
three sets of 6 cycle experiments were performed from 03 JST 5th. In addition to
the conventional data, GPS water vapor data and radar wind data were assimilated
every10min.
To reflect the analysis of the Inner LETKF in the Outer LETKF, the analyzed
value of the Outer LETKF was replaced by that of the Inner LETKF every 6 h at
the end of the assimilation windows of the Outer LETKF (namely, 09 JST and
15 JST of 5th), at which time both LETKFs produced the analyses. To reduce
the inconsistencies between the Inner LETKF and the Outer LETKF, the values
of the Outer LETKF at one and two grids inside from the boundary of the Inner
LETKF were produced by blending with those of the Outer LETKF (Fig. 20.4 b).
The weights used in blending are determined with linear interpolation.
If an Inner LETKF has to cover a wide area, huge computer resources are needed.
In this study, the wide area is divided into a number of small domains and the
Inner LETKFs are executed at each divided domain. When a number of Inner
LETKFs are deployed in the domain of the Outer LETKF, there might be overlapped
regions (Fig. 20.4 c). As the first step of the multi Inner LETKF, the values in the
overlapped regions were determined by averaging the analyzed values produced by
the aforementioned procedure. For instance, when N Inner LETKFs

80 80

.x i1 x iN /

were used, the values of the Outer LETKF ( x o / were obtained with the following

equation;

x o D . w o1 x o C w i1 x i1 C w o2 x o C w i2 x i2 CC w oN x o C w iN x iN /=N: (20.1)

This method allows the Inner LETKFs to be deployed flexibly. This method might
be more robust, because the replaced values of the Outer LETKF in the regions
where the domains of the Inner LETKFs are overlapped include the analyzed values
of the Outer LETKF.