Project: Advanced mathematical methods for the detection and attribution of non-stationarity in climate timeseries
Supervisors: Ted Shepherd, Jana de Wiljes and Antje Weisheimer
Project Description:
Anthropogenic climate change is increasingly apparent at the local scale, affecting extreme weather events and the seasonal cycle. Detecting and attributing those changes is of increasing importance for society. Yet there is considerable uncertainty at the local scale, including how regional climate change will unfold, and how those changes will translate to societal and environmental impacts.
From a mathematical perspective climate change is a form of non-stationarity: the systematic variation of the statistical properties of a distribution. Non-stationarity can arise from external forcings, but also from the low-frequency internal variability which can affect regional climate via atmospheric teleconnections. Climate models generally disagree on how these teleconnections will respond to external forcings, so the detection and attribution of their changes must allow for multiple possibilities. The statistical methods currently used in climate science do not allow for this. Causal networks/inference and storyline approaches offer new opportunities in this respect.
This project is designed to enhance the student’s proficiency in a state-of-the-art domain that integrates machine learning/artificial intelligence, climate data analysis, and uncertainty quantification. These skills are crucial for conducting sophisticated climate risk analysis on a regional scale, offering the student invaluable expertise in addressing complex environmental challenges.
The student will engage in hands-on experience with the latest model datasets provided by the European Centre for Medium-Range Weather Forecasts. They will innovate and design new algorithms aimed at detecting causal relationships, identifying change points that reveal hidden external influences, and capturing the intrinsic variability within the data.
