Abstract: Model errors and data errors are the limiting factors to the improvement of the numerical weatherprediction(NWP) models. In 4-dimension data assimilation, the model is often assumed to be precise, correctinginitial observations and neglecting the forecast error caused by the model error. Adjoint-model AssimilationSystem can be used in the correction of initial conditions and model parameters. Terrain is one of the multiplefactors governing atmospheric circulation and also an important ingredient of numerical weather prediction(NWP)models, responsible directly for the results. Model terrain adjustment is carried out by means of the PennsylvaniaState University(PSU)/National Center for Atmospheric Research(NCAR) non-hydrostatic mesoscale modelMM5 with its adjoint-model assimilation system(AMAS) including GMS5 cloud-derived winds andnon-conventional temperature from the Nation Satellite Meteorological Center(NSMC).There are twoconclusions:(1) the corrected terrain coordinates with the initial observations and the model can improve themodel results varying with the range and intensity of the meso-scale heavy-rain center. (2) Considering the effectof model errors on prediction, the application of AMAS is a new feasible approach to the improvement of NWPmodel parameters.