The Advances in Targeted Observations for Tropical Cyclone Prediction Based on Conditional Nonlinear Optimal Perturbation (CNOP) Method - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 582

both located in the core region of TC Longwang (2005) (Fig. 24.11 b, c), but the
accurate locations of the two maxima are slightly different: the location of CNOP
VIE maximum is one grid spacing to the south of CSV VIE maximum. Thus, the
CNOP Sen and CSV Sen are slightly different.
The CNOP Sen for TC Sinlaku (2008) is defined as a minimized square area,
large enough to contain the two VIE maxima of CNOP, and FSV Sen and CSV Sen
are another square area with the same size, capable of containing the two VIE
maxima of CSV and located on the north side of the storm center (Fig. 24.12 ).
In addition, the extents of CNOP Sen, CSV Sen, and FSV Sen for TC Longwang
(2005) are defined as a square area with

4 4 grid points, and that for TC Sinlaku

(2008) is defined as a square area with grid points.

In the four types of experiments (RA-CN, RA-CS, RA-FS, and RA-RA), 40
Ran Errs are generated by transforming forty

12 12

1 1;104 error vectors, which are

normally distributed, into forty 4 4 23 3 error matrices for TC Longwang (2005)

and by transforming 40 1 9;936 error vectors, which are normally distributed, into

forty error matrices for TC Sinlaku (2008). Specifically, the u-wind,

v-wind, and temperature initial states are perturbed, and the random error vectors are
zero means. Their standard deviations are 0.95 for TC Longwang (2005) and 0.32
for TC Sinlaku (2008).
Figure 24.13 show the nonlinear developments of a Ran Err from the CNOP Sen,
CSV Sen, and Ran Area, respectively, for both TCs. The development of the
Ran Err from FSV Sen is similar to that from CSV Sen for TC Longwang (2005)
since CSV Sen and FSV Sen for TC Sinlaku (2008) are located in the same position.
Obviously, as shown in Fig. 24.13 e, f, a rapid reduction in the magnitude of
wind and temperature of the evolved Ran Errs from Ran Area occur for both TC
Longwang (2005) and TC Sinlaku (2008), compared with the results of RA-CN
and RA-CS. For TC Longwang (2005), the verification dry energy norms of the
evolved Ran Errs from CNOP Sen, CSV Sen, and Ran Area are

12 12 23 3

Jkg 1 , 63:28

Jkg 1 and Jkg 1 , respectively, while for TC Sinlaku (2008) those values

17:47 12

Jkg 1 . Therefore, the locations of initial
errors in CNOP Sen may have had great impacts on the final forecasts.
Next, the impacts of 40 Ran Errs added to CNOP Sen, CSV Sen, FSV Sen,
and Ran Area are assessed according to the statistical averages of verification
dry energy norms of the 40 evolved Ran Errs from the four types of areas,
respectively. Table 24.2 presents the results of these statistical averages for both
TC Longwang (2005) and TC Sinlaku (2008). Notably, the errors introduced into
the CNOP Sen have the largest influences on the final forecasts. For both cases, the
Ran Errs introduced into the CSV Sen have the next largest influences, while the
Ran Errs fixed in Ran Area lead to the smallest changes. Thus, the growth rates
of Ran Errs introduced into sensitive areas, such as CNOP Sen and CSV Sen, are
higher than those introduced into Ran Area.
Additionally, another 36 randomly selected areas are considered in the
RA-RA experiment for TC Sinlaku (2008) and TC Longwang (2005). Similarly,
the statistical averages of nonlinear developments of the 40 Ran Errs from
the 36 local areas are obtained following the same procedure performed in

Jkg 1 , Jkg 1 ,and

are 21:13 14:10 3:08