Result shows that the performance of squall line simulation is very sensitive to model error associated with resolution and uncertainties in the physical parameterization schemes. At least a 10-km grid size is necessary to decently capture the squall line in this event. Relative to microphysics and PBL schemes, the simulated squall line is most sensitive to long-wave radiation parameterization schemes. For a grid size from 20 km to 5 km, using cumulus parameterization scheme degrades the squall line simulation relative to turning it off with a more severe degradation to grid size smaller than 10 km than coarser resolutions.
The sensitivity of the squall line simulation to the initial error was examined through probabilistic ensemble forecast. The performance of the ensemble simulation of the squall line is very sensitive to the initial error. About 15% of the ensemble members accurately capture the evolution of the squall line, 25% fail and 60% dislocate the squall line. Using different combinations of physical parameterization schemes for different members can apparently improve the probabilistic forecast. The lead time of this case is only a few hours. Error growth is clearly associated with moist convection development in terms of both time and space. A linear improvement in the performance of the squall line simulation is observed when the initial error is deceased gradually with the largest contribution from initial moisture field. This is different from the nonlinear improvement in the simulation of a squall line in the United States to the linear decrease of the initial error.