Master Thesis defense by Jonas Richard Damsgaard

Title:  Linking Arctic Warming to European Weather: Insights Through Data and Models

Abstract:
The rapid decline of Arctic sea ice and the amplified warming of the northern high latitudes have raised critical questions about the dynamical coupling between the Arctic and mid-latitude weather systems. In a warming climate observations and model simulations suggest that extreme, anomalouswinter conditions continue to impact Europe. A central point of debate is the extent to which these European cold extremes are driven by the retreat of sea ice in regions like the Barents-Kara (BK) Sea, versus internal atmospheric variability such as blocking events and meandering jet streams. This thesis investigates the occurrence, and shifting probability of extreme Cold Winter Months (CWMs) in Europe under the high-emission SSP585 forcing scenario. Specifically, this project intends to verify the results of Yang and Christensen 2012, which posits that CWMs will continue to occur at a roughly 50% rate in the medium-term future. This thesis utilizes a 57-member large ensemble of the EC-Earth3 climate model, allowing for a robust examination of how extreme winter anomalies can still manifest late into the 21st century.

The analysis is structured to quantify the changing probability of European CWMs over time and examine specific, late-century extreme events. The thesis evaluates the predictive power of the BK sea ice on European weather. This is initially achieved through conventional statistical approaches (Pearson correlation, 𝑅2, and RMSE) to test a simple linear relationship. Following this, machine learning techniques are applied, Standard Autoencoders (AE) and Variational Autoencoders (𝛽-VAE) are employed alongside XGBoost decision trees to extract and compress high-dimensional climate variables (geopotential height, surface temperature, latent heat flux, and sea ice concentration) across the Barents-Kara, North Atlantic, and greater Arctic regions. Through this framework, the study aims to determine the structural limits of predicting mid-latitude weather and assess the efficacy of dimensionality reduction in capturing the complex, non-linear dynamics of the climate system.

 Supervisor: Jens Hesselbjerg Christensen

Censor: Peter L. Langen