ACINN Graduate Seminar - SS 2025

2025-03-26 at 12:00 (on-line and on-site)

Introducing mrCOSTS: Unsupervised learning of coherent spatiotemporal structures

Karl Lapo

ACINN, University of Innsbruck, Austria

 

The unsupervised and principled diagnosis of multi-scale data is a fundamental obstacle in earth sciences. Existing methods, such as modal analysis, time frequency analysis, data-driven modeling like Dynamic Mode Decomposition (DMD), and even deep learning are not well-suited to diagnosing multi-scale data, usually requiring supervised strategies such as human intervention, extensive tuning, or selection of ideal time periods. We present the multi-resolution Coherent Spatio-Temporal Scale Separation (mrCOSTS), a data-driven method explicitly designed for the challenges of multi-scale data. It is a hierarchical variant of DMD that enables the unsupervised diagnoses of spatiotemporal features by decomposing the data into bands of temporal dynamics associated with coherent spatial modes.

We demonstrate mrCOSTS on: 1) a complex toy system, 2) Antarctic sea ice concentration, and 3) the mountain boundary layer. In each example we demonstrate how mrCOSTS can be used to trivially recover a robust diagnosis of complex dynamics. For the sea ice concentration we diagnose the relative contribution of interannual variability and climate change to recent extreme events. For the mountain boundary layer example we demonstrate mrCOSTS on LIDAR data from both the CROSSINN and TEAMx pre-campaigns. We demonstrate a scale-aware evaluation of a numerical weather prediction model against LIDAR observations and present a generalization of Reynold's decomposition driven by mrCOSTS.

 

 

The content of this webpage is the author’s opinion and the author’s intellectual property. ACINN has the author’s consent to make this information available on the webpage for the announcement of the seminar presentation and in the seminar archive.