Master Thesis Defence by Eskil Nordahl Fundal

Title: A Self-Organizing Map Perspective on 75 Years of North Atlantic Atmospheric Circulation Variability and Teleconnections to Surface Climate

Abstract: 
The North Atlantic atmospheric circulation exerts great influence on European regional climate, and it is proposed both theoretically and by climate models that anthropogenic climate change may affect the North Atlantic atmospheric circulation. Therefore understanding its variability is crucial for understanding future European regional climate change. 

This thesis investigates the atmospheric circulation variability in the North Atlantic and its impact on seasonal mean climate in Europe for the period 1950-2024 using the unsupervised machine learning method 'Self-Organizing Maps'. The method is used to obtain a season-wise non-linear daily classification of the atmospheric conditions in the North Atlantic in the period based on ERA5 daily mean sea level pressure. From this classification the changes in the frequency of occurrence of the mapped atmospheric circulation patterns are compared using three baseline periods; 1950-1974,1975-1999 and 2000-2024. In addition, the impact of these frequency changes on the seasonal mean climate in Europe is calculated using a statistical framework.

The analysis identifies that in spring, summer and autumn there have only been minor atmospheric circulation changes with only limited impact on seasonal mean climate in Europe, while for winter there have been changes affecting around 10% of the winter season. These dynamic changes have contributed significantly to regional climate changes, particularly on the Iberian Peninsula and in Northern Europe, but a large part of European temperature, precipitation and wind changes in the analysis period are still dominated by thermodynamic effects. In addition, the identified atmospheric circulation patterns for the winter are used to explore the continuum of the North Atlantic Oscillation (NAO) to illustrate the heterogenous, non-linear and non-stationary nature of the NAO and its teleconnections. Lastly, the thesis demonstrates a purpose-driven and quantitative approach for training and evaluation of Self-Organizing Maps for atmospheric circulation analyses, addressing a major methodological gap in the existing literature.

 Supervisor: Jens Hesselbjerg Christensen