The impact of optimal selected historical forecasting samples on hybrid ensemble-variational data assimilation

A resampling method that selects historical forecasting samples as supplementary samples is proposed for the hybrid ensemble-variational data assimilation system to alleviate the computational burden of ensemble forecasting samples. To select reasonable samples from all historical forecasting samples, the first modes of absolute vorticity are abstracted by the empirical orthogonal function (EOF) as indicators of atmospheric dynamic features from the background and each of historical forecasting sample, then they are matched at the analysis time. A series of single observation tests and 19-day cycling data assimilation and forecasting experiments for a Mei-yu period are carried out to evaluate the impact of the selected historical forecasting samples. The single observation tests indicate that the use of selected historical forecasting samples is able to provide reasonable flow-dependent background error covariance for the data assimilation system. The cycling data assimilation and forecasting experiments demonstrate that the analyses and forecasts as well as precipitation forecast skills are improved by using the combination of selected historical forecasting samples and ensemble forecasting samples. The sample-combined experiment performs close to the experiment with full-size ensemble forecasting samples, but it spends fewer computational resources. The diagnosis of a heavy rainfall case is presented to further illustrate the role of the selected historical forecasting samples. It is found that the simulation of vertical velocity and relative humidity are improved for the case in the experiment of the combined samples, leading to better intensity and position forecasts of the precipitation.