A scientist from the University of Oxford has developed a new computer algorithm which can be applied to Earth System Models to reduce spin-up time.
Funded by the Agile Initiative, the scientist tested the models used in UN Intergovernmental Panel on Climate Change (IPCC) simulations.
The new computer algorithm was shown to be, on average, ten times faster at spinning up the model than currently-used approaches.
The time taken to reach a stable equilibrium, which is needed for a simulation to run, was reduced from many months to under a week.
Study author Samar Khatiwala, Professor of Earth Sciences at the University of Oxford’s Department of Earth Sciences, who devised the algorithm, said: “Minimising model drift at a much lower cost in time, and energy is critical for climate change simulations, but the greatest value of this research may ultimately be to policymakers who need to know how reliable climate projections are.”
The work is published in the journal Science Advances.
What are the current issues with Earth System Models?
Earth System Models, essential for forecasting climate change, simulate how Earth’s components interact. They predict future extreme weather and climate events, informing initiatives like the IPCC reports.
Yet, a key challenge for climate modellers is ensuring model stability. Before simulating the climate after the Industrial Revolution, models must reach equilibrium pre-industrial states. This ‘spin-up’ phase prevents erroneous attributions to human factors.
However, this process is slow, demanding thousands of model years and up to two years on supercomputers for IPCC simulations.
About the new computer algorithm
Professor Khatiwala’s new algorithm uses a mathematical method called sequence acceleration. This equation was later applied by DG Anderson in the 1960s to speed up solutions for Schrödinger’s equation.
With over half of the world’s supercomputing capacity focused on solving this problem, ‘Anderson Acceleration’ is now a widely used algorithm for it.
Recognising the iterative nature of both problems, Khatiwala found that Anderson Acceleration could also be used to shorten model spin-up times.
Using Anderson’s scheme, the final solution is attained more quickly by combining previous outputs into a single input.
As well as making the spin-up process much faster, the concept can be applied to a huge variety of Earth System Models that are used to investigate and inform policy on a number of climate-related issues.
Professor Khatiwala is now working with a number of research groups, including the UK Met Office, to trial the new algorithm in their climate models.
