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 576

schemesA,B,C,E,F,andG,theresultofschemeHismoresimilartothatof
scheme E than to those of schemes A, B, C, F, and G. This indicates that the areas
north of the initial cyclone (scheme E) have a significant influence on the results.
Therefore, verifications area should not be too large, or the results are affected by
some irrelevant areas.
In general, the design of the verification area is important in tropical cyclone
targeted observations. To improve a tropical cyclone forecast with targeted observa-
tions, the verification area must include the tropical cyclone tracks during the rele-
vant time period. To do this, the ensemble forecast results must be consulted, which
can introduce uncertainty into the prediction. In addition, the background field, the
potential sensitive areas, as well as the economically relevant area can also provide
references for the design of the verification area. The verification area cannot be
too large or small, as the results would be affected for the former and the important
information would be missed for the latter, also it is hard to catch the best positions
of the cyclone for the latter. However, when the CNOP method is used to identify
the sensitive areas, once the general position of the verification area is determined,
a small variation in its size or position has minimal influence on the identification
of the sensitive areas. This is a favorable characteristic for targeted observations.

24.3.4 Sensitivity of CNOP Sensitive Areas with Respect

to the Optimization Period

The optimization period is a key issue in the choice of the cases. It is sure that
the similarities between sensitive areas are rare for cases that occur in completely
different situations; therefore, considerable attention has been paid to similarities
between sensitive areas during cases occurring under similar conditions or during
the temporal evolution of individual cases. The study of Zhou and Mu ( 2012b )
focused on the latter. They have studied the following two tropical cyclones: Matsa
(2005) and Meari (2004).
A set of experiments has been designed to study the time dependence of the
sensitive areas in the context of targeted observations. Except for the studied time
period, all parameters are held constant throughout the set of experiments. Two
approaches are used. In the first approach, the initial time is fixed and forecasts
are generated for 12, 24, and 36 h later. The initial times are set at 1200 UTC 4 Aug
2005 for the Matsa case (Fig. 24.4 a) and 1200 UTC 25 Sep 2004 for the Meari case.
In the second approach, the forecast time is fixed and forecasts are generated from
initial times 12, 24, and 36 h prior to the forecast time. The forecast times are set at
0000 UTC 6 Aug 2005 for the Matsa case and 0000 UTC 27 Sep 2004 for the Meari
case. The three initial times are therefore 1200 UTC 5 Aug 2005, 0000 UTC 5 Aug
2005, and 1200 UTC 4 Aug 2005 for the Matsa case (Fig. 24.4 b), and 1200 UTC 26
Sep 2004, 0000 UTC 26 Sep 2004, and 1200 UTC 25 Sep 2004 for the Meari case.
The optimisation time periods are the same as the forecast time periods in this study.
First, the nonlinearity of the typhoon cases is explored by comparing the linear
FSVs and nonlinear CNOPs. Results suggest that for both two approaches, the