The background error variance is modified within the Japan Meteorological Agency (JMA) global four-dimensional variational (4D-Var) data assimilation system; the impact is investigated. In the operational 4D-Var the background error variance is assumed to be constant globally at each vertical level for each variable. This study performs additional data assimilation experiments by allowing horizontal and temporal inhomogeneities in the background error variance. A new method with a similar idea to the one for the dynamic quality control is proposed in this study to account for temporal variations, i.e., flow-dependence. It turns out that the impact by the horizontal and temporal inhomogeneities is small and that including the horizontal variations does not necessarily improve the forecast. However, considering flow-dependence indicates generally positive impact.