Representation of Model Error in Convective-Scale Data Assimilation: Additive Noise Based on Model Truncation Error

To account for model error on multiple scales in convective-scale data assimilation, we incorporate the small-scale additive noise based on random samples of model truncation error and combine it with the large-scale additive noise based on random samples from global climatological atmospheric background error covariance. A series of experiments have been executed in the framework of the operational Kilometre-scale ENsemble Data Assimilation system of the Deutscher Wetterdienst for a 2-week period with different types of synoptic forcing of convection (i.e., strong or weak forcing). It is shown that the combination of large- and small-scale additive noise is better than the application of large-scale noise only. The specific increase in the background ensemble spread during data assimilation enhances the quality of short-term 6-hr precipitation forecasts. The improvement is especially significant during the weak forcing period, since the small-scale additive noise increases the small-scale variability which may favor occurrence of convection. It is also shown that additional perturbation of vertical velocity can further advance the performance of combination.