In an attempt to produce accurate simulations of the initial soil moisture state, forcing data sets for land surface models have been assembled through bias correction of the ECMWF and NCEP/NCAR reanalysis. The methodology of the bias correction ensures that mean monthly 2-meter air and dew point temperatures, short and longwave downwelling radiation, and precipitation within the corrected products match global observations. The two corrected data sets were used as forcing for land surface model simulations of soil water content. Comparisons between the soil water fields produced by each simulation demonstrate that despite the identical bias correction, a given soil moisture initialization will contain a large amount variance. From these comparisons, masks of initial soil moisture uncertainty have been created to illustrate its seasonal and geographical extent, how these variations change with depth, and time frames over which these variations persist.