Quantifying climate projection uncertainty arising from the signal-to-noise paradox

Background

Making atmospheric predictions beyond short-term weather forecasts requires predictable signals that change the likelihood of particular climate events amidst the unpredictable atmospheric noise. In seasonal climate prediction, numerical models are used to forecast the evolution of the atmosphere and ocean for the next few months. These models produce forecasts of North Atlantic climate that, on average, better predict the real world than themselves. Given that models and initial conditions are imperfect, we would expect the opposite. This has become known as the signal-to-noise paradox (SNP). The SNP is interpreted as an underestimation of the predictable signal in models, but its root cause remains a mystery. A critical open question is whether the problem also affects multidecadal projections of future climate – the basis for decision-making on climate change. This project seeks to establish if and how known signal-to-noise errors translate into uncertainty in climate projections.

PhD opportunity

The project aims to reduce uncertainty in climate projections by developing novel statistical approaches to connect model errors across timescales and to recalibrate projections to remove the effects of signal-to-noise errors. It will seek to identify metrics of the SNP that can be used to constrain future projections. An example of this kind of approach is Smith et al (2025), which adjusted North Atlantic Oscillation projections to account for an overly weak simulated response to volcanic eruptions. This project would expand this work to consider potentially weak responses to increasing greenhouse gases.

This PhD is aligned to a large NERC funded project involving the Universities of Exeter and Leeds, the National Oceanography Centre and Met Office. The successful applicant would join an expert team and contribute to addressing one or more of the following scientific challenges:
1. Advance the detection and diagnosis of the SNP across timescales
2: Quantify projection uncertainty arising from the SNP
3: Develop statistical approaches to recalibrate projections for the SNP
4: Use machine learning to develop and test calibration approaches

Broadly, the project will compare observations to climate predictions, to seek to quantify and reduce uncertainty in climate projections related to the SNP. There are opportunities to analyse new large ensembles of climate model simulations, including at high resolution, and novel AI (machine learning) seasonal forecasts. Within this broad context, the student will be supported to co-design a programme of research that best suits their skills, training needs, and career vision, whilst addressing questions of scientific and societal importance and with the potential for long-lasting impact. In all cases, the project will provide opportunities to develop collaborations with partners, including the Met Office, and provide training in numerical modelling, statistics and data science, and weather and climate physics.

Applicant Profile

The ideal applicant will possess a strong background in a mathematical or physical science and interest in weather and climate science.

Other information

https://www.nature.com/articles/s41612-018-0038-4
https://www.nature.com/articles/s41586-020-2525-0
https://www.nature.com/articles/s41558-025-02277-2