Application Of Physical Filter Initialization Four-dimensional Variational Data Assimilation Technique To Precipitation Nowcasting Of Numerical Model

Nowcasting,in terms of the weather forecast of 0 to 3 hours,is mainly based on radar echo or satellite image extrapolation methods.Timeliness,accuracy and high-efficiency are the most prominent feature of nowcasting,especially for the rainstorm and other high impact weather.As one of the main bases of disaster warning,nowcasting is an important means for mitigating the hazard caused by weather.However,the prediction ability of extrapolation methods generally drops rapidly within 3 hours,because of the lack of the physical mechanism of the occurrence,development and extinction of convective weather systems.While the prediction ability of numerical model gradually rises within 3 hours,the nowcasting system can be developed based on the numerical model if the data assimilation technique can produce more accurate initial conditions.The PFI-4DVar assimilation method(Physical filter initialization Four-dimensional variational data assimilation)has the ability to make the analysis fields physically and dynamically balanced with the numerical model using the filter in the process of assimilation.It is helpful to short the adjust process during model integration,thus shorting the spin-up time of the model and getting better nowcasting skills.Based on the weather research and forecasting model(WRF)and its data assimilation system(WRFDA),in this paper,six hour forecast field from the National Centers for environmental prediction(NCEP)Global Forecast System is selected as background fields(with horizontal resolution of 1°× 1°).The model domain is with center at 37°N and115°E,horizontal resolution of 9km,150 × 150 grid points,and 50 vertical levels.In order to reduce the demand of computing resources,a 12 minute assimilation time window is chosen.Then the 3-hour nowcasting can be completed in a limited time to ensure the PFI-4DVar method can be used in the numerical prediction of nowcasting.The CV7 option of background error covariance is chosen which is with independent U and V components in WRFDA and "gen_be" module based on NMC(National Meteorological Center)method is used to generate background error covariance matrix.The TS(threat score)score of 6-hour accumulated precipitation in different parameter settings was compared,and the best variancefactor var_scaling and scale factor len_scaling are set to be 1.5 and 0.05,respectively.Eleven weighting coefficients(0 to 1)were tested,and the best value of weighting coefficient is chosen to be 4×10-6 according to the TS scores of the rainfall forecasts of the experiments.The performance of PFI-4DVar is examined using a rainfall case in North China on August 11,2018.The results of PFI-4DVar experiments with data assimilation are compared with those of the experiment without data assimilation.The ETS(equitable thread)scores of every 1 hour accumulated rainfall forecast of the PFI-4DVar data assimilation experiment are improved obviously with in the 6 hours.Especially,the performance of precipitation forecast in the 1 to 3 hours is improved dramatically compared with the experiment without data assimilation.The precipitation can be predicted with distributions close to the observations in the first hour.The data assimilation processes can be finished within 30 minutes,so it is possible to be sued in nowcasting.In the experiment with PFI-4DVar,assimilating of ground and sounding observations changes the initial water vapor field.Meanwhile,assimilating of radar radial winds changes the dynamic field.Through data assimilation,the humidity in the lower layer of the precipitation area is increased,and the convergence of the lower layer and the divergence of the upper layer are enhanced.Thereafter,the precipitation forecasts are improved.In order to further analyze the main reasons of the improving of the prediction in the data assimilation experiment,the precipitation of the model are separated in grid resolvable precipitation and sub-grid precipitation.According to the 6 hours accumulated precipitation and 1 hour accumulated precipitation in first 3 hours,the total precipitation are mainly contributed from the sub-grid precipitation.The sub-grid precipitation has a great influence on the precipitation nowcasting in this experiment.In the short time range,the precipitation is predicted by cumulus convection parameterization process.It is found that the cumulus convection parameterization process is a fast response mechanism in the short time forecast,which can produce precipitation at the initial time of the model forecast.The experiments based on 17 precipitation cases of north China in August,2018 are also carried out in this study.The results show that the PFI-4DVvar technique can significantly improve the prediction skills of the heavy rainfall.Data assimilation increases the ETS scores of the heavy rainfall from 0.125 to 0.190(50% raised)with the 6 hours accumulated precipitation greater than 25.0 mm,and from 0.016 to 0.081(4 times increased)for the threshold of 60.0 mm.Moreover,in the 17 experiments,cumulus convective parameterized precipitation still has a larger impact on the short time precipitation forecasts.At the same time,it can be seen that assimilation observation data has also improved the grid resolvable precipitation forecast,especially improved the forecast of the intensity of the precipitation center.In addition,in order to compare and discuss the influence of assimilation methods and data on numerical forecast,the experiments based on the precipitation case on August 11,2018 are also carried out in this study using different data assimilation methods with different observation data.The 6-hour accumulated precipitation forecasts are carried out and the results are compared.The results show that the four-dimensional variational method can predict the detailed distribution of precipitation by assimilating radar data,and PFI-4DVar has the best performance for the precipitation distribution.To sum up,this study uses the physical filter initialization four-dimensional variational assimilation technology(PFI-4DVar)to carry out the nowcasting experiments.The ability of PFI-4DVar for improving precipitation nowcasting is explored.The results show that PFI-4DVar assimilation method can significantly improve the skills of precipitation nowcasting.Assimilating observations is effective for improving the grid resolvable precipitation,but most of the precipitation comes from the sub-grid precipitation,which are produced by the cumulus convection parameterization scheme.The results of multi-case experiments show that the PFI-4DVar data assimilation method can predict the distributions of the precipitation within very short integrating time.