In the ‘90s, Krishnamurti's group at The Florida State University introduced physical initialization to improve precipitation predictions in the tropics produced by the atmospheric general circulation model. More lately, physical initialization was extended to regional spectral models, with applications initially in South American precipitation forecasting. Physical initialization combines physical and dynamical procedures. During the physical procedure, a numerical weather prediction model assimilates “observed” precipitation fields from day -1 to day 0, the latter usually represents the forecast initial time. To weakly constrain predicted fields to the “observed” large-scale features, Newtonian relaxation towards global analyses is also applied to all prognostic variables, except specific humidity. This technique has been used in the initialization of primitive equations-based models for weather prediction, and listed among the early precipitation assimilation schemes. Over the years, the physical initialization procedure has changed to better simulate regional hydroclimatology via continuous assimilation of precipitation information. Examples of physical initialization applications in South and North American summertime climate simulations will be presented at the meeting, together with new insights on this still challenging topic of precipitation assimilation in regional models.