Time-hierarchical Clustering and Visualization of Weather Forecast Ensembles

Florian FerstlMathias KanzlerMarc Rautenhaus and Rüdiger Westermann

Department of Informatics, Technische Universität München, Germany

Abstract

We propose a new approach for analyzing the temporal growth of the uncertainty in ensembles of weather forecasts which are started from perturbed but similar initial conditions. As an alternative to traditional approaches in meteorology, which use juxtaposition and animation of spaghetti plots of iso-contours, we make use of contour clustering and provide means to encode forecast dynamics and spread in one single visualization. Based on a given ensemble clustering in a specified time window, we merge clusters in time-reversed order to indicate when and where forecast trajectories start to diverge. We present and compare different visualizations of the resulting time-hierarchical grouping, including space-time surfaces built by connecting cluster representatives over time, and stacked contour variability plots. We demonstrate the effectiveness of our visual encodings with forecast examples of the European Centre for Medium-Range Weather Forecasts, which convey the evolution of specific features in the data as well as the temporally increasing spatial variability.

Associated publications

Time-hierarchical Clustering and Visualization of Weather Forecast Ensembles F. Ferstl, M. Kanzler, M. Rautenhaus, R. Westermann, IEEE Transactions on Visualization and Computer Graphics 2017 (Proc. IEEE SciVis 2016) [PDF] [BIBTEX]