Short-range Precipitation Forecasting with Real-time 4DVAR Analysis Using Dense Radar and Surface Observation Data

A four-dimensional variational Doppler radar analysis system (VDRAS) is implemented for severe storm cases occurred in Tokyo metropolitan area, Japan. Radial velocity data from single C-band and triple X-band radars and reflectivity factor data from C-band radar are assimilated into the first-guess which is produced by merging with interpolation of dense surface observations and analysis/forecast field of the Weather Research and Forecast (WRF) model. Experiments are first carried out to examine the sensitivity of the retrieved and subsequently forecasted fields with respect to system configurations. The use of cycling 4DVAR, in which sequential data assimilation is implemented twice, improves forecasting precipitation than the use of cold-start 4DVAR, and horizontal length-scale in recursive filter is relatively sensitive in case of the cycling configuration. In the cold-start configuration, the weights assigned to the spatial smoothness constraints are relatively sensitive to the analysis fields. Constants of evaporation process in warm rain microphysics and the raindrop fallspeed have much effect on results.

The performance of short-range precipitation forecasting with VDRAS analysis is investigated. Results indicate that an anelastic non-hydrostatic model in VDRAS works well for 2-hour nowcasting of the evolution of convections in this study. The limitation of lead times depends on the domain size and computation time. The domain is small (100 km x 100 km) because the outside of Tokyo area is mountainous and the application of VDRAS is favorable to flat areas. Such a small domain is easy affected by the boundary condition with the lapse of time. Basically, VDRAS has much capability in the real-time nowcasting in virtue of introducing Message Passing Interface (MPI). Some small-scale storms, however, need a higher horizontal resolution less than about 2 km, which causes difficulties for the real-time operation. We will also discuss on a framework of assimilating VDRAS analysis into WRF model analysis/forecast as one of remedies to extend the lead times up to around 6 hours. In some storm cases mainly due to the convergence of water vapor flux, the accurate water vapor analysis is crucial for the accurate retrieval. The usefulness of GPS precipitable water data from the Japanese dense GPS receiver network (GEONET) is discussed to modify the first-guess in the pre-processor of VDRAS as a final discussion. The combined use of data from different observing systems is promising to improve quantitativeness of precipitation forecasting.