Diurnally varying background error covariances estimated in RMAPS-ST and their impacts on operational implementations

Background error covariance (BEC) plays a key role in variational data assimilation systems. The National Meteorological Center (NMC) method has been used widely to generate forecast error samples for BEC estimation. At present, most variational-based rapid update and cycling (RUC) data assimilation and forecasting systems use a fixed BEC without consideration of diurnal variation. In this study, diurnal variation BECs were estimated using three month forecast error samples (0000 UTC 01 June to 2100 UTC 31 August 2019), which came from the Rapid-refresh Multi-scale Analysis and Prediction System-Short Term (RMAPS-ST). Series of single observation tests and one-month partial cycling data assimilation and forecasting experiments with diurnal variation BECs were carried out based on the system. The results showed the following: 1) Diurnal variation is found in the standard deviation of forecast error samples, with the minimum value of standard deviation appearing at nightfall (0900 UTC, 1700 BJT), and the maximum value appearing at the early morning (2100 UTC, 0500 BJT). The eigenvalues also show similar diurnal variation features, indicating the diurnal variation characteristics of background error are consistent in the physical space and the EOF space. 2) The diurnal variation of BEC is further verified by single observation tests, and the analysis increments well response to the BEC diurnal variation. 3) The results of one-month cycling experiments show that diurnal variation BECs could improve the assimilation and forecasting performance of RMAPS-ST.