Clustering numerical weather prediction ensemble output is an excellent way to group the ensemble output data and easily assess future possible weather scenarios. It is a technique that bridges the gap between purely visualizing ensemble model output either by displaying several dozens maps or in hard to discern ‘spaghetti maps' and the back-end statistics such as means, percentiles, uncertainty bands, and probabilities. The latter format is particularly useful for explicit risk forecasting in planning purposes, however, the ‘numbers' give no feel for the underlying forecast information or weather regime while the former format can be too much information for a user to adequately discern weather regimes.

Clustering sorts the model forecasts and presents the data in an organized way. In this presentation the clustering method at MeteoGroup will be presented. We run the clustering method for both the ECMWF model as well as the GFS model. The clustering is performed for the 500hhPa height field and applied to other parameters. We will discuss the advantages and disadvantages of clustering, compare results of various methods, and show some examples of a ‘grand ensemble cluster' - ECMWF and GFS model combined into a 70 member ensemble. Additionally, output from combining ensemble data from various initial model start times will also be discussed with the following question in mind, “Will the members of both models be spread out over all clusters, or will they organize themselves? “