Assimilation of radar reflectivity volumes through a LETKF scheme for a high-resolution numerical weather prediction model

Despite the continuous improvements in numerical weather prediction (NWP) models, the quantitative precipitation forecast (QPF) is still a challenge. A crucial role in the accuracy of QPF is played by data assimilation, the technique whereby initial conditions for an NWP model are generated by combining observations of the state of the atmosphere and a previous forecast of the model itself. In this work, the direct assimilation of radar reflectivity volumes, which is still at a preliminary stage in operational frameworks, is evaluated. This is carried out using a local ensemble transform Kalman filter (LETKF) scheme and employing the COSMO-2I model, the configuration of the convection-permitting model of the COnsortium for Small-scale MOdelling (COSMO) adopted at the Regional Agency for Prevention, Environment and Energy of Emilia-Romagna region (ARPAE) to provide high-resolution weather forecasts over Italy. The crucial aspects of the assimilation of this type of observation are investigated, in particular concerning the length of the assimilation window, the estimation of the observation error and the configuration of the radar operator employed to simulate reflectivity observations from the prognostic model fields. Taking advantage of the results obtained from this investigation, a set-up for the direct assimilation of reflectivity volumes suitable for an operational implementation is defined. Accuracy of QPF and of other forecast variables obtained with this set-up is compared to that obtained with the current operational set-up employed at ARPAE to generate the initial conditions of COSMO-2I, in which radar-estimated precipitation is assimilated through a latent heat nudging scheme. Results of this comparison, which is the most extended ever performed in terms of number of forecasts involved and in the number of verification scores employed, suggest that time is ripe to directly assimilate reflectivity volumes in an operational framework using an ensemble Kalman filter scheme.