Observation impact in a convective‐scale localized ensemble transform Kalman filter

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.