Exploratory analysis of climate data using source separation methods

The goal of this research project is to use novel machine learning techniques such as independent component analysis (ICA) and more general denoising source separation (DSS) for extracting significant patterns of global climate variability.

Contents

• NCEP/NCAR reanalysis data
• Demonstration of DSS for extracting El Niño
• ENSO as the component with the most prominent interannual variability
• Most prominent interannual phenomena with seasonally changing effects
• Frequency-based representation of the slowest climate variability
• Weather signals with structured variance
• Groups of dynamically coupled components with predictable dynamics
• Our publications and related literature

Group members

• Alexander Ilin
• Harri Valpola
• Erkki Oja
• Jaakko Luttinen (research assistant)