C9 - Visual feature analysis from individual cases to collections of ensembles

Project leaders: Dr. Marc Rautenhaus, Prof. Dr. George Craig

Other researcher: Andreas Beckert (PhD student), Christoph Fischer (PostDoc)

Meteorological studies that target three-dimensional features in the atmosphere (e.g., jet streams and fronts), their mutual relation, their uncertainty as represented by ensemble simulations, and their influence on predictability are encountered in a number of projects within W2W. Recent research, including our work in Projects A7 and B5 in Phase 1, has suggested that to perform such studies, novel visualization techniques based on 3D feature detection and interactive visual analysis (IVA) workflows have the potential to greatly facilitate the analyzes and to reveal information that could easily be missed with conventional 2D techniques. However, for analysis tasks as proposed in Phase 2 projects, suitable methods are still lacking.

This project builds on our work in Projects A7 and B5 in Phase 1 and proposes investigation into new visual analysis techniques suitable for specific tasks put forward by selected meteorological W2W projects, closely collaborating with them. Of particular interest is to create a bridge from analysis of individual cases to analysis of entire collections of ensembles of similar cases (e.g., multiple cases of different cyclones that are of similar type) to obtain statistical information. We will follow a twofold approach targeting feature-based analysis (where an atmospheric feature is considered a spatially separable 3D structure that evolves with time, e.g., a frontal surface) and index-based ensemble analysis (where an index is a scalar measure of a characteristic property of a feature or atmospheric state, e.g., the average temperature on a frontal surface). This will allow us to first study individual cases of a given dynamical phenomenon using detailed 3D visualization, then to extend the analysis to an ensemble simulation and further to collections of ensembles in order to identify characteristics of the phenomenon that can be linked to predictability. Expressed in a generalized way, we will look into techniques that detect and jointly visualize not only single but multiple types of 3D features and, in particular, their mutual relation in complex dynamical situations. This will then enable the scientist to abstract relevant feature and feature relation characteristics into index quantities that subsequently can be linked to predictability and be analyzed across ensembles. Specific phenomena that we will consider include intense winter storms, the tropopause, and stratosphere-troposphere coupling. Challenges will be posed by complex dynamical scenarios to which existing feature-based techniques need to be extended and for which new detection schemes need to be developed. Of particular interest will be mutual feature relation, e.g., to classify a jet core line depending on its relation to a frontal surface. Finally, an IVA workflow will be developed to analyze ensemble variability based on indices derived from features and their relations and, in particular, to identify links to forecast skill.
This project differs from the other visualization projects A7 and B5 in its explicitly interdisciplinary character. It will be a major effort in linking visualization and meteorological activities for the benefit of both fields: our work will contribute to the feature-based and ensemble visualization literature to advance the state of the art in visualization, at the same time the meteorological community will obtain new visual analysis techniques to investigate their research questions.