<p>One strategy to mitigate sampling errors and model errors for ensemble-based data assimilation methods is covariance inflation. An additive inflation method is proposed to the current China Meteorological Administration (CMA) operational global data assimilation (DA) system. It uses an ensemble four-dimensional variational (En4DVar) system that incorporates flow-dependent background error covariances from short-term ensemble forecasts, and a static multiplicative inflation method to increase ensemble spread. Compared to the traditionally applied static multiplicative inflation, the proposed additive inflation has dynamical error growth and can better consider the impact of model errors on flow-dependent background error covariances. Single-point observation experiments indicate that additive inflation retains flow-dependent characteristics of the analysis increments, and produces smaller magnitudes of analysis increments but more accurate analyses than multiplicative inflation. Cycling assimilation experiments demonstrate that compared to multiplicative inflation, additive inflation improves the analyses of geopotential height and temperature than multiplicative inflation, and the advantages of additive inflation over multiplicative inflation are mainly contributed from small wavenumbers. Forecasts launched from analyses indicate that additive inflation produces significantly improved forecasts than multiplicative inflation for all state variables from surface to 10 hPa both over the northern and southern hemispheres, although the significant improvements over the southern hemisphere have smaller magnitudes and shorter lead times than over the northern hemisphere.</p>