AN INITIALIZATION METHOD TO INCORPORATE PRECIPITATION DATA INTO A MESOSCALE NUMERICAL WEATHER PREDICTION MODEL

An initialization method for the Japan spectral model (JSM) is developed for the use of precipitation observations from a radar-raingauge network over Japan. The initialization executed the physical initialization method to calibrate thermodynamic and dynamic variables of objective analysis data in such a way that model precipitation from the calibrated data is equal to precipitation observations. Afterwards, the non-linear normal mode initialization with the precipitation process, which was similar to that developed by Takano and Segami (1992), was conducted. Forecast experiments on a case of Baiu-frontal disturbances (June 30, 1990) show: 1) The initialization method reduces both spin-up error and position error of precipitation forecasts. 2) Reduction of position error is ascribed to the modification of relative humidity due to calibration of thermodynamic variables, while, in the model precipitation region where no precipitation was observed, this calibration adopted a primitive method which reduces relative humidity to an empirically-determined threshold value. 3) Calibration of dynamic variables contributes greatly to determining the initial p-velocity. This resulted in improvements in precipitation forecasts for the first few hours. 4) The non-linear normal mode initialization with the precipitation process itself made little change to the initial p-velocity, even in the strong model precipitation areas, while Takano and Segami (1992) pointed out that this procedure did not largely change the initial divergence field due to the weak diabatic heating. Since the non-linear normal mode initialization with the precipitation process retains circulations which are accompanied with precipitation, this procedure contributes to the reduction of slow spin-up of precipitation caused by adiabatic normal mode initialization.