Continuous assimilation of lightning data using time-lagged ensembles for a convection-allowing numerical weather prediction model

In this study, a lightning data assimilation method based on the time-lagged ensembles for predicting severe convection is presented. With the lightning data assimilation scheme, the background error covariances are computed using time-lagged ensembles, which consist of deterministic forecasts from eight forecast cycles initialized every 3hr. Pseudo-observations of graupel mixing ratio (q(g)) are retrieved from total lightning rates by utilizing empirical vertical profiles obtained from the simulation results of the previous forecast cycles, and the corresponding observation errors are estimated according to the uncertainties in the lightning observations and the empirical vertical profiles of q(g). The increments of the model state variables are computed with the Kalman gain matrices and are continuously ingested into the Weather Research and Forecasting model via nudging terms acting on the prognostic equations over each time step during model integration. The effect of the lightning data assimilation scheme on convection analysis and forecast was assessed through a case study of a severe convective event, which took place in the Guangdong of China. Assimilating lightning data recovered many of the observed convective cells, suppressed the spurious convection, and corrected the displacement errors of the convective systems. Quantitative verifications indicate that forecast skills were improved mainly in the convective regions with the impact of assimilating lightning data on stratiform regions being overall less effective.