Mesoscale Modelling for Tropical Cyclone Forecasting over the North Indian Ocean - Monitoring and Prediction of Tropical Cyclones in the Indian Ocean and Climate Change - page 275

Fig. 4: Mean (of all 24 cases) vector displacement errors (VDEs in km) in
12-hr interval for CNTL and 3DVAR experiments.
simulated constituting 16 cases. A set of three numerical experiments are
conducted for each case at 9 km horizontal resolution. The first experiment
(CNTL) uses the NCEP FNL analyses for the initial and boundary conditions.
In the second experiment (GTS), all the Global Telecommunication System
(GTS) observations are assimilated into FNL initial conditions for an improved
analysis. The third experiment (DWR) uses the DWR radial velocity and
reflectivity along with GTS observations. Figure 5 provides the distribution of
GTS and DWR data for Jal cyclone at 00 UTC 6 November 2010.
The DWR data assimilation experiment showed a significant improvement
in the short-term forecast of outer spiral bands of hydrometeors (Qrain, Qcloud
and Qvapor) around the storm centre. The realistic evolution of surface latent
heat flux and the improved prediction of hydrometeors help better evolution of
intensity in the case of DWR data assimilation experiments. The DWR
experiment shows better improvement of track prediction.
Figure 6 shows improved tracks (shown one case for each cyclone) to be a
result of improved initial vortex position, upper-level divergence field and
steering current. The better evolution of 850 hPa relative vorticity is also another
reason for better track prediction. The mean vector displacement errors (VDEs
in km) of all 16 cyclone cases for the three experiments are calculated and
presented in Fig. 6(e). It is clear that the mean VDEs are less for the DWR
experiment tracks which varies between 50 and 250 km from the 12-hr forecast
to the 72-hour forecast. The VDEs of the GTS experiment ranges up to 400 km
while in the CNTL, the errors are even higher and up to 500 km. The gain in
skill of the DWR and GTS experiments with respect to the CNTL experiment
clearly demonstrates that the skill of the DWR data assimilation is significantly
high when compared to that of the GTS experiment, in particular, for longer