The WRF (Weather Research and Forecasting) model has been used to study an impact of different soil moisture initialization schemes on a performance of local weather forecasting over the region covering the SE States. Three types of soil moisture (SM) initialization have been tested: (1) SM produced by ETA operational mesoscale model and considered as a control set; (2) SM interpolated from NARR (North American Regional Reanalysis); (3) SM fields generated by the NOAH LSM available from LIS (Land Information System) at 5-km resolution. Before performing case studies simulations (for the Summer and Fall 2005), the 8-year spin-up integration period of the NOAH model was used to generate realistic distributions of the soil moisture content over the region. Real-data meteorological forcing fields from the NLDAS archive has been used for this spin-up. Sensitivity of a daily magnitude of the surface air temperature and the water vapor content to various types of SM initialization will be evaluated for typical cases of wet and dry SM conditions at regional scale. Also, impact of assimilation of the MODIS land surface temperature into WRF initial analysis scheme on a local weather forecasting performance will be examined.