Berlin 2012 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 29: Posters II

DY 29.43: Poster

Thursday, March 29, 2012, 17:00–19:00, Poster A

Comparison of paired and independent forecast ensembles by univariate and multivariate skill measures — •Stefan Siegert and Holger Kantz — Max-Planck-Institut für Physik komplexer Systeme, Dresden, Germany

A forecast ensemble is a collection of runs of a numerical model. The ensemble members are constructed by adding perturbations to the observed initial state (the analysis). In numerical weather prediction, the spread of such a forecast ensemble is used to estimate the uncertainty of the numerical model about the future state of the atmosphere.

The generation of perturbations for weather forecast ensembles is non-trivial and computationally expensive. A cheap method to increase K, the total number of ensemble members, without having to calculate further perturbations is the construction of pairs of ensemble members, by adding and subtracting the same perturbation from the analysis. However, such an ensemble can only span a (K/2)-dimensional subspace of the model space, whereas a fully independent ensemble can, in principle, span a K-dimensional subspace.

We study how well paired and fully independent ensembles are able to represent the variability of the verifying observation. We analyze their skill on a univariate (gridpoint-wise) basis, using the outlier statistic and the continuously ranked probability score. Further, ensemble variability is studied in a multivariate sense, using minimum spanning tree analysis. We find systematic differences between the two kinds of forecast ensembles.