Ensemble-based Singular Value Decomposition Analysis for a Local Rainfall Event Predicted by Water Vapor Assimilation

To improve the accuracy of forecasts of local rainfalls by numerical models, it is important to assimilate dense observations (e.g., Radar reflectivity) based on the relationship between the hydrometeors (e.g., rain, snow, and graupel) in the rainfall regions and the atmospheric state (e.g., wind, temperature, and water vapor) around the precipitating area. A statistical analysis based on ensemble forecasts with large members using a cloud-resolving model is one of the promising approaches to investigate such relationship and observations, which are essential for the improvement of the analyzed state.

To this end, we conducted 301-member ensemble forecasts of a local rainfall with the Japan Meteorological Agency Non-Hydrostatic Model (JMANHM, Saito et al. 2006) of 1-km horizontal grid spacing. This rainfall event occurred on the Kanto Plain in Japan at about 1430 JST (Japan standard time) on 4 August 2016. Its horizontal scale and lifetime were less than 10 km and about 1 hour, respectively. Since it occurred within 10-km distance from the Meteorological Research Institute (MRI) in the Tsukuba City, the hydrometeors and the atmospheric state in and around the rainfall were well observed by the MRI C-band polarimetric radar, the water vapor Raman lidar of the MRI, Global Navigation Satellite System (GNSS), and so on. We assimilated these dense observational data with the 300-member local ensemble transform Kalman filter (LETKF, Hunt et al. 2007) of 1-km horizontal grid spacing, and conducted 301-member ensemble forecasts using the initial states given by 300-member LETKF analyses and their average.

The assimilation of water vapor data obtained by the GNSS and the Raman lidar turned out to hasten the genesis of the rainfall. Using these ensemble forecasts, we performed an ensemble-based singular value decomposition analysis of the covariance between the hydrometeors and the atmospheric variables. It revealed that the hydrometeors in the rainfall were mainly correlated with the atmospheric state only below the 2-km height before the rainfall although they were correlated with the atmospheric state from the ground to over 10-km height after the rainfall. In particular, water vapor of the northern and southwestern sides of the Tsukuba City near the surface was highly correlated with the hydrometeors. This result indicates that the hydrometeor data assimilation is important for correcting the atmospheric state such as water vapor and its convergence near the surface.