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.

Over the past 15 years, adjoint-based, ensemble-based and hybrid methods have been developed for estimating observation impact based on the forecast sensitivity to observation impact (FSOI). These methods are now commonly used in global modelling systems. However, little attention has been given to assessing observation impact in regional convection-permitting modelling systems. This study presents the first evaluation of ensemble-based estimates of observation impact over an extended period of six weeks in such a convection-permitting modelling system, namely the regional ensemble system of Deutscher Wetterdienst. Another aspect that has received little attention is the choice of the forecast-error verification metric. Nearly all previous studies used the difference between the forecast and a subsequent analysis of the same modelling system expressed in terms of energy (total energy norm). While such a self-verification generally needs to be treated with caution, it appears unsuitable for convection-permitting regional forecasts. Firstly, total energy does not really reflect parameters that forecast users are interested in, and important forecast quantities such as surface wind gusts and precipitation are not even part of the analysis. Secondly, systematic analysis and forecast errors are non-negligible in the presence of convection, especially for important variables that are related to convection. To overcome this issue, we introduce the use of independent radar observations for the verification of observation impact and compare results for a variety of different observation-based metrics for a six-week high-impact weather period in summer 2016. This revealed a particular sensitivity of the estimated impact to model as well as observation biases and sensitivity studies indicated that even small biases can have an influence on the estimated impact. Additionally, we demonstrate that FSOI can be used to identify biases through comparison of results for different metrics.

Knowledge on the contribution of observations to forecast accuracy is crucial for the refinement of observing and data assimilation systems. Several recent publications highlighted the benefits of efficiently approximating this observation impact using adjoint methods or ensembles. This study proposes a modification of an existing method for computing observation impact in an ensemble-based data assimilation and forecasting system and applies the method to a pre-operational, convective-scale regional modelling environment. Instead of the analysis, the modified approach uses observation-based verification metrics to mitigate the effect of correlation between the forecast and its verification norm. Furthermore, a peculiar property in the distribution of individual observation impact values is used to define a reliability indicator for the accuracy of the impact approximation. Applying this method to a 3-day test period shows that a well-defined observation impact value can be approximated for most observation types and the reliability indicator successfully depicts where results are not significant.

The aim of the present study is the accuracy and sensitivity assessment of a recently developed approximation method for observation impact, i.e. the contribution of observations to forecast-error reduction. The considered method uses an analysis and forecast ensemble for the approximation and does not require the adjoint model. The method is implemented for the first time in a convective-scale limited-area modelling system and its accuracy is assessed through comparison with results from a number of data denial experiments. It has been found that the difference from data denial is not significant and it is possible to assess the impact of subgroups of observations and detect disadvantageous or improperly used observations.

Data assimilation algorithms require an accurate estimate of the uncertainty of the prior (background) field that cannot be adequately represented by the ensemble of numerical model simulations. Partially, this is due to the sampling error that arises from the use of a small number of ensemble members to represent the background‐error covariance. It is also partially a consequence of the fact that the geophysical model does not represent its own error. Several mechanisms have been introduced so far to alleviate the detrimental effects of misrepresented ensemble covariances, allowing for the successful implementation of ensemble data assimilation techniques for atmospheric dynamics. One of the established approaches is additive inflation, which consists of perturbing each ensemble member with a sample from a given distribution. This results in a fixed rank of the effective model‐error covariance matrix. In this article, a more flexible approach is introduced, where the model error samples are treated as additional synthetic ensemble members, which are used in the update step of data assimilation but are not forecast. This way, the rank of the model‐error covariance matrix can be chosen independently of the ensemble. The effect of this altered additive inflation method on the performance of the filter is analyzed here in an idealized experiment. It is shown that the additional synthetic ensemble members can make it feasible to achieve convergence in an otherwise divergent parameter setting of data assimilation. The use of this method also allows for a less stringent localization radius.

No abstract

No abstract

No abstract