AI mapping between drivers of future climate uncertainties and weather and climate impacts in Africa

Full Description

Background

There is an urgent need for reliable assessments of future high-impact weather under climate change.

This is true globally, but Sub-Saharan Africa is not only highly vulnerable, but presents unique challenges for climate prediction.

Moreover, African climate planning typically focus on 5-30 year timescales, where we need to account for interactions between climate change and multi-yearvariability.

Tropical weather is dominated by convection, which is poorly captured in global models.

Now, convection-permitting models (CPMs) can provide improved predictions of extremes, often revealing greater increases.

CPMs are, however, computationally costly and represent a limited sample of potential climate change uncertainties, with almost no assessment of how this interacts with climate variability.

Machine-learning provides an opportunity to capture the added value of CPMs, to expose how high impact weather may change, accounting for uncertainties in natural variability and climate change.

PhD Opportunity

How do we explore how interactions of climate variability and change are likely to impact Africa?

We have rich information on variability and change from global model experiments, but lack the impact on Africa.

The Met Office has a new world-leading ensemble of high-resolution CPMs for Africa that explicitly model the rain-generating convective storms that especially important for extremes.

However, their vast computational cost has limited these simulations to sample only a few multi-year datasets that are insufficient to address interactions of variability and climate change.

This project will develop machine-learning (ML) approaches to downscale global models.

The CPMs over Africa provide data to train ML to relate large-scale circulation (that is modelled in current climate projections) to local scale weather (that usually requires the CPMs to model).

Objectives

1 Incorporate physical understanding on the drivers of tropical weather into the development, benchmarking, and evaluation of key phenomena for tropical weather features key to climate risk, including mesoscale convective systems (MCSs).

2 Use these benchmarks to train and improve ML downscaling for Africa.

3 By establishing skill of ML to capture high-resolution climate, use ML to provide locally impact relevant information on African weather and climate that would be expected to arise from both climate variability and climate change on 10-30 year time horizons.

This will enable a first assessment of the relative impacts on local hazards, arising from major modes of climate variability and climate change, respectively.

This will inform our physical process understanding and allow well-founded narratives of possible future risk, including from rare events.

The student will be working alongside a new Gates-funded project developing similar approaches, but for application to sub-seasonal predictions.

This will provide opportunities for collaboration in the UK and Africa.

Applicant Profile

The project will provide exciting opportunities for students with a strong background in maths, physics, statistics, computer science or meteorology, who want to develop machine-learning tools to provide improved projections.

There may be opportunities to travel to Africa, but this is not essential. Visits to the UK Met Office will facilitate collaboration with the Met Office co-supervisor, as well potentially with other scientists there.