The improvement in the forecast skill of quantitative precipitation forecasts (QPFs), especially during summer season, depends critically on the accuracy of model physical parameterizations and the quality of model initial conditions. With a rapid increase in the amount of observations from remote-sensing instruments, % and the advanced techniques for directly linking the prediction models with observations, a mesoscale data assimilation %, at a higher resolution than that of global models and with model physics describing more accurately the clouds and rainy systems, may be needed to make better use of these observations for improved 48-h QPF on smaller scales. In this study, hourly multi-sensor rainfall data from the National Centers for Environmental Prediction/Climate Prediction Center (NCEP/CPC) will be used for both data assimilation and evaluation of the model physics. These data are available at a 4-km resolution and are generated by combining approximately 3000 automated hourly raingage observations available over the contiguous 48 states with the radar precipitation estimates from the Next Generation Weather Radar (NEXRAD) network. Ideas of using rain observations to adjust empirically temperature and moisture profiles in physical initialization are tested for the effective use of rain observations in a variational framework using adjoint techniques. Uncertainties in the model parameters which are used in the physical parameterizations are assessed through a set of least-square fit experiments minimizing a cost function measuring the distance between model simulations and rain observations. A digital filter is included as a weak constraint in the 4D-Var procedure to remove the high frequency gravity wave oscillations that could be created by the assimilation of rain observations. %inconsistency between the initial guess fields and the forecast models, as well as the assimilation of rain observations. The effectiveness of the proposed method for the use of satellite and radar derived precipitation estimates and raingage observations will be demonstrated on the prediction of a few selected severe weather cases.