Multimodality in weather prediction

To deal with the inherent uncertainties in weather and climate predictions, the common practice is to apply ensemble methods to produce a probability distribution of variables of interest. Informatively summarizing such probability distributions is a non-trivial task and the commonly used means and standard deviations result in the loss of crucial information, especially in the case of multimodal distributions with distinct likely outcomes. In this project, we aim at machine learning and visualization methods to summarize results from ensemble models in a physically more meaningful and conceptually more intuitive way, while at the same time preserving significant information contained within the ensemble.