B5 - Data-driven ensemble visualization

Principal investigator: Prof. Dr. Rüdiger Westermann

Other researcher: Alexander Kumpf (PhD)

In a number of research projects in Research Area B, numerical simulations of meteorological processes are carried out using systematically or stochastically varied physical parameters and initial conditions. Major goals are the exploration of the sensitivity of meteorological phenomena to these variations, the analysis of how different processes, and resulting weather events, depend on each other, and the usage of these simulations to form correlation structures for data assimilation. The numerical simulations generate an ensembles of atmospheric states, where each ensemble member shows a possible occurrence of different physical processes.

In this project we will develop visualization techniques for three-dimensional (3D) ensembles to enable an improved (statistical) exploration of such fields. We will provide means to interactively visualize individual ensemble members and their temporal variations, including approaches to track regions exhibiting certain properties such as specific dust or cloud droplet concentrations.

Highly efficient techniques to compute statistical quantities from an ensemble will be embedded into the visualization process, to analyze the variations between ensemble members as well as statistical relationships between data values at different locations and time points. Further investigations will address the grouping of spatial locations and ensemble members regarding their statistical similarities, and the analysis of the dynamics of selected clusters.

We will also investigate means to effectively relate the particular occurrences of weather events in the simulation with the input parameterizations which were used to generate these simulations. Efforts in this direction can eventually help to enable improved parameter space navigation (i.e. to find the set of parameterizations which cover the most probable outcomes) and to hint towards specific properties and effects of the used initial conditions, physical models and parameters.

To achieve the specific contributions, we will develop interactive ensemble visualization techniques using statistical data measures, so-called location-based ensemble visualization, rather than making our approaches dependent on specific meteorological features derived from the data. Thus, we expect many of the algorithmic, statistical, and conceptual aspects we address, such as efficient data structures and computational schemes, visual mappings of statistical properties and similarity metrics for them, to be directly transferable to other Research Areas where ensemble exploration is required. In addition, we will work closely together with selected projects to address the question which statistical properties are most suited to support ensemble exploration in specific subfields of meteorology.

Our activities can be grouped into the following main areas:

• development of a scalable high-performance infrastructure for statistical data processing and 3D visualization
• development of similarity metrics, and clustering approaches using these metrics, for spatial fields and locations in these fields
• improvement of ensemble sensitivity analysis for assessing the dependencies of forecast variables to initial state variables
• development of graphical representations for statistical features and clusters, and their dynamics
• development of user interfaces to support intuitive selection and analysis of interesting data characteristics