Concepts for a pattern-oriented analysis ensemble based on observational uncertainties

The idea is proposed of an analysis ensemble of deterministic, model-independent analyses. The ensemble is based on random perturbations of irregularly distributed observations. The purpose of implementing an analysis ensemble is to define uncertainties in analysis fields due to their observational background and errors. As one possible application, the uncertainty information could, in future, be used to define confidence intervals for verification measures depending on the reference data. The analysis system VERA and a high-resolution Central European observation network are used as a testbed for the development of the methodology. Several approaches for defining weights for the perturbation fields are investigated and compared. Basic weights are determined by a sophisticated data quality control scheme producing error estimates for observations. These estimates can be combined with additional information attempting to include more explicitly the spatial structure of the observed fields in the ensemble. The information is provided by either a principal component analysis of a time series of analysis fields or a 2D-discrete wavelet transform. Strengths and weaknesses of the different adjustments for ensemble analysis perturbations are discussed. It is shown that perturbations provided by the wavelet-based approach lead to useful results for several meteorological parameters tested.