Exploring uncertainty due to clouds in modelled future climate change systematically

Changes in clouds are one of the biggest uncertainties affecting predictions of future climate change. Because their size is typically much smaller than the gridlength of numerical climate models, clouds must be “parametrised”, meaning that they are represented approximately using information from larger-scale conditions.

It is difficult to quantify the difference between climate models, which typically have different model parametrisations written in terms of different functions. Some progress has been made through “perturbed physics ensembles”, which take one model structure and perturb uncertain model parameters through their ranges of possible values. However, climate models are expensive to run, meaning that only a few parameter combinations can be tried. Features of model behaviour for unexplored parameter combinations must instead be estimated via statistical emulation. Parameter values that are unrealistic given available data are then identified via “history matching”.

We have developed Continuous Structural Parametrisation (CSP), which is a way of approximating structurally different model parametrisations as functions of the same variables, effectively writing them as members of a perturbed physics ensemble. CSP also allows us to represent observations or high-resolution process models within the same structure, allowing us to benchmark our parametrisations.

Clouds are the most uncertain process affecting the interaction between radiation and climate, playing a key role in the degree to which the planet will warm in future. This project will start by writing down cloud parametrisations and high-resolution model realisations of cloud using CSP. The research can then take a number of directions:

1) A Gaussian process emulator or neural network can be applied in conjunction with CSP to provide a more complete description of a parametrised process while still retaining human-readability. Our preliminary work indicates that this is helpful for cloud, but more statistical modelling work is needed to constrain low cloud, which is notoriously

difficult to understand.

2) Emulation can fill in the CSP parameter space, giving us a prediction of key cloud responses between the parametrisations that we have. History matching can then be used to find which parts of our parameter space are not inconsistent with high-resolution models and observations, and highlight new areas of parameter space that model parametrisations should explore, allowing new climate models to better explore future climate impacts.

3) The CSP emulator can be placed directly within the Met Office Unified Model climate model to explore the potential impacts of uncertainty in cloud parametrisation on future surface climate.

The project is linked to the UK Met Office with collaborator Mark Webb providing expertise in clouds and climate modelling. Co-supervisor James Salter is a statistician who will provide expertise in emulation and history matching.

This project will suit students with a strong background in mathematics or statistics or another numerate degree such as Physics or Meteorology, and an interest in Earth’s atmosphere, climate and climate change. Experience in coding in python or another high-level language would be an advantage, but is not essential.