The impact of assimilating observed precipitation in the Regional Spectral Model (RSM) as part of a Physical Initialization (PI) procedure is being investigated. We currently nudge atmospheric moisture, since the atmospheric humidity adjustment leads to stable temperature and wind changes and the RSM precipitation then remains close to the observed precipitation (80-90% correlation). Experiments nudging the atmospheric heating are also being developed to determine if further improvements can be made. In particular, RSM simulations over Brazil and the US, forced only by the large-scale global analysis, are being compared to simulations that also incorporate PI for various convection schemes (Kuo, SAS, RAS). Unfortunately, this increased simulation skill does not necessarily result in increased long-range forecast skill. However, additional long-range forecast skill may result from coupled interactions of a more realistic precipitation with the RSM land surface, which may thus be better initialized by this PI than it is currently initialized by the global analyses. Preliminary RSM predictability experiments using PI are also underway and will be reported on in a companion study.