There are two main sources of estimate uncertainty in Numerical Weather Prediction (NWP), i.e., initial‐value‐related uncertainty, model‐related uncertainty. A 20‐member mesoscale ensemble forecasting system including these two kinds of uncertainty is set up to simulate tropical cyclone Danny 1997. The track of ensemble means is the best compared with all 20 members after 12 h. The sensitivity of simulation result to the two kinds of uncertainty is studied. The result shows that both kinds of uncertainty make an important role in simulation, but they dominate in the different time period during the simulation. The initial value‐related uncertainty dominates the simulation during 0~12 h of model integration. The model‐related uncertainty grows steadily during model integration. High sensitive area of uncertainty is located where severe weather occurs around the center of cyclone.

This paper presents a new horizontal staggered grid (LE grid), which defines geopotential height h at a gridpoint, and both u and v at the same mid-gridpoint along the x and y directions. A general method is used to deduce the dispersion relationships describing Rossby waves on LE grid and Arakawa A-E grids, which are then compared with the analytical solution (AS) in resolved or unresolved cases, using the two-order central difference or four-order compact difference scheme in the frequency and group velocity. The results show that in both resolved and unresolved cases, no matter whether two-order central difference or four-order compact difference scheme is used, the frequency and group velocity discrete errors on LE grid in describing Rossby waves are smaller than those of Arakawa A-E grids. At the same time, for the LE grid and Arakawa A-E grids the employment of a four-order compact difference scheme with higher difference precision can not inevitably improve their accuracy of frequency and group velocity in x and y directions in describing Rossby waves.