Master Thesis defense by Marc Peradalta I Negre
Title: Large-scale causal drivers of winter precipitation in Europe and the North Atlantic.
Abstract:
Large-scale atmospheric dynamics, responsible for a significant share of winter precipitation in the Euro-Atlantic sector, remain only partially understood despite their relevance for seasonal forecasting and climate change projections. This thesis challenges the standard approach of characterising precipitation through the North Atlantic Oscillation (NAO), arguing that it acts as a statistical descriptor integrating the effects of many remote mechanisms rather than being a driver itself.
By means of a causal discovery framework, direct large-scale drivers of precipitation are successfully identified at every grid point, with dependencies found to be predominantly linear at monthly timescales. Non-trivial causal links are uncovered, including a connection between Barents--Kara sea ice and Iberian Peninsula precipitation that correlation methods fail to detect, as well as evidence that the stratospheric polar vortex directly modulates precipitation across both northern and southern Europe. These drivers naturally organise local precipitation into coherent regions sharing the same causal structure, resolving most of the Euro-Atlantic domain. Some areas, however, remain unresolved, notably the transition zone between the two NAO poles. To assess the predictive power of a physically grounded characterisation, the identified drivers are used as predictors in a regression framework. Both linear and non-linear models yield forecast skill that outperforms the NAO in spatial coverage and explained variance, showing that understanding the mechanisms driving precipitation also improves the ability to predict it.
The results point to several promising directions for future research, particularly the extension to submonthly timescales and indirect teleconnection pathways, where synoptic variability and upstream mechanisms may refine the identified causal signals.
Supervisor: Jens Hesselbjerg Christensen, Ramiro Ignacio Saurral, BSC (Barcelona Supercomputing Center).
Censor: Peter L. Langen, Aarhus University.