A Feature-Based Framework to Investigate Atmospheric Predictability

• Published in: Monthly weather review Vol. 153; no. 8; pp. 1431 - 1450
• Main Authors: Schmidt, Sören, Riemer, Michael, de Heuvel, Jorge, McTaggart-Cowan, Ron, Selz, Tobias
• Format: Journal Article
• Language: English
• Published: Washington American Meteorological Society 01.08.2025
• Subjects: Atmospheric conditions; Automation; Case studies; Coherence; Experiments; Forecast errors; Simulation; Spatial analysis; Spatial filtering; Statistical analysis; Statistical methods
• ISSN: 0027-0644; 1520-0493
• DOI: 10.1175/MWR-D-24-0090.1

The flow dependence of atmospheric predictability implies that forecast errors grow more rapidly in some atmospheric conditions than in others. A better understanding of this flow dependence thus requires a local analysis of error growth. To facilitate such an analysis, this study introduces a feature-based perspective. While feature identification and tracking is often applied to atmospheric systems, associated forecast errors exhibit small-scale structure and thus lack spatial coherence. Consequently, using a standard feature approach, merging and splitting of features are ubiquitous, which severely limit the ability to automatically identify distinct temporal feature evolutions and subject them to statistical analysis. While the spatial filtering of data alleviates this inherent challenge, it does not resolve it and incurs a loss of information. We overcome this challenge by introducing a feature postprocessing that combines individual features into regional-scale entities, which exhibit much increased spatial and temporal coherence. It is these postprocessed entities that prove suitable for subsequent feature-based analysis. We demonstrate the utility of the feature-based perspective by applying it to the spread of global ensemble experiments designed to assess upscale error growth. Analyses are exemplified that contribute to an improved understanding of the flow dependence of error growth mechanisms and that link error growth characteristics to local atmospheric conditions.