Project: The signal-to-noise problem in weather and climate forecasts
Supervisors: Jochen Bröcker (UoR), Eviatar Bach (UoR) & Antje Weisheimer (ECMWF)
Project Description:
Operational Weather and Climate Forecasts are probabilistic, meaning that the forecast not only indicates a range of possible events but also the associated uncertainties. This implies that the forecast can predict its own eventual accuracy. A puzzling phenomenon, called the “Signal-to-Noise Paradox (SNP)”, has been observed especially in seasonal forecasts: the reality appears to be more predictable than the forecasts are suggesting through their self-assessing nature. This project will work towards understanding the SNP better. Tools to diagnose the SNP in Weather and Climate forecasts will be developed. Furthermore, dynamical mechanisms will be identified (in the context of medium complexity models of geophysical fluid dynamics) that could possibly give rise to the SNP.
Specifically,
- Combine approaches from Machine Learning and modern statistics to handle large weather and climate data sets to diagnose the SNP in seasonal and climate forecasts.
- Combine approaches from Machine Learning with ECMWF seasonal forecasts to increase the ensemble size and estimate the forecast signal with higher accuracy.
- Explore physical insights of Climate science, to identify dynamical mechanisms that lead to the SNP.
- Apply and test the developed methodologies in the context of medium complexity models of geophysical fluid dynamics.
