In order to test the forecast capability of MM5, seven combinations of three different physical parameterization schemes are used in simulation tests of precipitation prediction during 2005-flood season over the low latitude plateau. The results show that the combinations of Goddard microphysics parameterization, Grell cumulus parameterization and cloud/RRTM radiation parameterization in the third nested grids are superior to the others, especially the forecast of downfall and rainstorm. There is a certain influence for different scanning radius of interpolation prediction to the observational network. The interpolation result is comparatively ideal when the 20—30km scan radius is used.