Numerical weather prediction models require accurate initial conditions and therefore rely on the quality of the assimilation scheme. The radiance measurements from satellites have been most successfully assimilated into the global weather prediction systems under clear atmospheric conditions. To fully utilize the information of satellite measurements under all weather conditions for numerical weather prediction, we need to enhance the data assimilation scheme by introducing the variational retrieval of cloud water profile. This study implements the forward model as discussed in Part I (Forward and Ajoint Models) under the main title to one-dimensional variational retrieval (1DVAR) scheme to simultaneously retrieve the surface and atmospheric parameters using AMSU data. Retrieved parameters include surface temperature, emissivity, and profiles of temperature, water vapor and cloud water liquid. NCEP global data assimilation system (GDAS) daily products are used to produce the initial guesses for atmospheric temperature, humidity and cloud liquid water profiles. However, the initial guess of surface emissivity is computed using our previously developed microwave emissivity model with GDAS surface parameters (e.g. soil moisture, skin temperature, canopy water, snow depth). It is shown that the quality of retrieved atmospheric and surface parameters is much improved, compared with the separate retrievals.