New algorithm could lead to better predictions of future climate change

Simulations with Earth System Models can take months, or even years, to run. The process requires a so-called “spin-up” phase up-front to reach a stable equilibrium (after taking into account potentially thousands of years of real-world conditions) before results can aid in predicting future climate change. Time spent running these models on supercomputers is extremely expensive, so any additional time it takes can make these costs prohibitive, but without the spin-up phase the model can “drift”, simulating changes that might wrongly be attributed to manmade factors.

However, a new algorithm devised by Agile Initiative Sprint researcher Professor Samar Khatiwala could significantly decrease the time-sink in the spin-up process. In a paper published in Science Advances, Professor Khatiwala describes that this new algorithm is on average 10 times faster at spinning up typical models than currently-used approaches, which could reduce the time from months to less than a week.

These models form the foundation for predictions of future extreme weather and climate event scenarios, so the impact of devising a much quicker algorithm could be very significant. Professor Khatiwala said, “perhaps the greatest value of this research may ultimately be to policy makers who need to know how reliable climate projections are.”

Professor Khatiwala leads the Sprint team “Can we solve the ‘spin-up’ problem in Earth Systems Modelling?”, which brings together expertise from the Department of Earth Sciences and the Mathematical Institute, with fellow researchers Professor Coralia Cartis and Andrew Wilson.

To ensure broad adoption of this new method for spin-up it will be important to demonstrate its applicability to many different ocean and land models. After such promising results as were reported in this paper, the sprint team has had interest from several climate modelling centres around the world. The climate change simulations these centres perform are the basis for the IPCC reports, and with the next one due in 2029 they are already starting to spin-up their models.

The team are working with a number of them to trial the new algorithm and software. The UK Met Office is one of them and, as the external partner for this sprint, has been closely involved from the outset. The National Center for Atmospheric Research (NCAR) in the USA is another; their Community Earth System Model (CESM) is one of the most widely used in the field of climate science. The sprint team has already begun work on applying the method to the land component of CESM with exciting early results.

To learn more about the research, visit the Sprint page, or read the piece by Professor Khatiwala in The Conversation. And to read the paper, visit Science Advances: https://www.science.org/doi/10.1126/sciadv.adn2839