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 577

Fig. 24.4 The design of the optimization time periods for TC Matsa (2005). ( a )forthefirst
approach and ( b ) for the second approach (From Zhou and Mu 2012b )
nonlinearity in Matsa case is strong, especially at longer forecast integrations, since
the CNOPs become progressively more different from the FSVs as the forecast time
extends further from the initial time (Fig. 24.5 , for the first approach, figure for the
second approach not shown). While the Meari case is weak nonlinearity regardless
of the optimisation time period as the patterns of the CNOPs and FSVs are similar
(Fig. 24.6 , for the first approach, figure for the second approach not shown). So the
Meari case will be interchangeably called the “linear case” and the Matsa case the
“nonlinear case” for the remainder of this part.
For the first approach, the comparison of the sensitive areas identified for the
Matsa and Meari cases reveals several interesting features. In the linear case,
the sensitive areas identified for a special forecast time are consistent with those
identified for other forecast times when the initial time is fixed (Fig. 24.7 ). This
result means that targeted observations deployed to improve a special time forecast
would also favourably affect the forecasts at other times. In the nonlinear case,
however, although there are some similarities in the sensitive areas identified for
different forecast times, these similarities are limited (Fig. 24.8 ). This indicates
that although the targeted observations deployed for a special time forecast are
also beneficial for other times' forecasts, the forecast improvements for other
times are limited. So it is suggested that in the nonlinear case, the deployment
of targeted observations should be adaptive to obtain the largest improvement for
different targeted forecasts, and should be more widespread to achieve the greatest
improvement in multiple time forecasts. In addition, for both two cases, the closer
the forecast times, the higher the similarities of the sensitive areas.