Abstract
Abstract. A hybrid variational ensemble data assimilation has been developed on top of the HIRLAM variational data assimilation. It provides the possibility of applying a flow-dependent background error covariance model during the data assimilation at the same time as full rank characteristics of the variational data assimilation are preserved. The hybrid formulation is based on an augmentation of the assimilation control variable with localised weights to be assigned to a set of ensemble member perturbations (deviations from the ensemble mean). The flow-dependency of the hybrid assimilation is demonstrated in single simulated observation impact studies and the improved performance of the hybrid assimilation in comparison with pure 3-dimensional variational as well as pure ensemble assimilation is also proven in real observation assimilation experiments. The performance of the hybrid assimilation is comparable to the performance of the 4-dimensional variational data assimilation. The sensitivity to various parameters of the hybrid assimilation scheme and the sensitivity to the applied ensemble generation techniques are also examined. In particular, the inclusion of ensemble perturbations with a lagged validity time has been examined with encouraging results.
Highlights
Data assimilation is the process of utilising meteorological observations to determine the initial state for Numerical Weather Prediction (NWP) models
We compare the impact of the ensemble generation technique on the quality of the hybrid assimilation and we examine the sensitivity of the hybrid assimilation scheme to various parameters
The forecast model used in the experiments was the HIRLAM grid point forecast model
Summary
Data assimilation is the process of utilising meteorological observations to determine the initial state for Numerical Weather Prediction (NWP) models. The total assimilation increment δx can be considered to include two parts, one part δxvar corresponding to the constraint given by the climatological background error covariance, and another part that is a linear combination, with space-dependent weights, of the ensemble perturbations, i.e. the deviations between the ensemble members and the ensemble mean: K δx = δxvar + κ αk. For the ensemble constraint Jens we need to specify the general weight βvar of the climatological background error constraint, the corresponding weight βens βvar βvar − 1 of the ensemble constraint, the variance of the ensemble weighting function αk, and the horizontal correlation spectrum of the ensemble weighting function αk These variables can be considered as tuning coefficients of the hybrid variational ensemble data assimilation.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
