Numerical weather predition is strongly dependent on the accuracy of initial conditions. Data assimilation has been proved to be a major advance in NWP during past decades. The work described in this paper is an attempt to explore the impacts of the data assimilation from nonconventional observations in mesoscale models, such as MM5 and COAMPS, based on a convective initiation database for the central Gulf Coasts for the summer of 1996. Previous numerical simulations of these cases, based on only "cold start" for both models, failed to correctly simulate the observed precipitations. For the control experiment, MM5 overestimated the rainfall and failed to simulate the diurnal variation of surface temperature while COAMPS underestimated the observed rainfall. The results show the improvements of precipitation, surface temperature, water vapor pressure and wind forecast after applying a data assimilation scheme for both models based on multivariate optimal interpolation (MVOI) scheme and incremental update, by which radiosonde, surface observations and nonconventional data are assimilated.