Abstract: Based on the methods of WRF (Weather Research and Forecasting Model),WRF-3DVAR (Weather Research and Forecasting-Three-Dimensional Variational Data Assimilation) and the NMC (US National Meteorological Center),we investigated whether or not and how the statistical background error samples under different initialization schemes have non-neglected impacts upon the background error covariance and its related forecasting effects.Experiments with hot-start cycled assimilation and cold-start initialization schemes were carried out.It is seen that the results from single point observation idealized assimilation experiments with background error covariance calculated from statistical background error samples under different initialization schemes are consistent with those from theoretical models.The differences in analyzed incremental fields are obvious and should not be ignored.The experimental results of "Soudelor" typhoon rainfall event in 2015 indicate that the background error covariance calculated from background error samples under different initialization schemes has a certain effect on the assimilation fields,and exerts obvious effects on the forecasts of typhoon track and rainfall,but have little impacts on the typhoon intensity and maximum wind speed near the typhoon center.Numerical experiments with four different initial time indicate that the background error covariance calculated from background error samples under hot-start cycled assimilation schemes shows a much better forecast for "Soudelor" typhoon track and rainfall,and its impacts on typhoon track are mainly shown after 24-h integration of the numerical forecast model.

Key words: Numerical forecast, Background error covariance, Background error, Typhoon