In a new study, Oxford Mathematician Coralia Cartis and Samar Khatiwala from Oxford's Department of Earth Sciences, together with colleagues and support from across the UK, Europe and the USA, have developed a novel approach to speed up the optimisation of ocean biogeochemical models - critical tools for predicting the impacts of climate change on marine ecosystems and the global carbon cycle.
Ocean biogeochemical models simulate how nutrients, carbon, and oxygen flow through the ocean. These models are essential for understanding how marine ecosystems respond to climate change and how the ocean regulates carbon dioxide levels in the atmosphere. However, these models are often too computationally expensive to systematically optimise due to the need for long “spin-up” periods - simulations that allow the model to stabilise before real testing begins.
To carry out their research, the team used transport matrices extracted from the MIT General Circulation Model (MITgcm). By incorporating their shortened spin-up method, they demonstrated that the optimised model’s performance remained robust, even with reduced simulation time. The high efficiency of using transport matrices from the MITgcm made it the perfect tool to test and validate their optimisation method.
“Spin-up times are a major bottleneck when calibrating ocean biogeochemical models to observations,” says Sophy Oliver, a climate scientist at the UK’s National Oceanography Centre. “By showing these times can be reduced, we can refine and calibrate our models faster, which should ultimately lead to more reliable projections of how marine ecosystems react to climate change.”
The team tested their method by applying it to a commonly used ocean biogeochemical model, speeding up its spin-up phase from 3000 years down to 2000 years. The results showed that even with a shorter spin-up it was possible to reliably optimise the model, a result that the authors believe is broadly applicable to other ocean models.
This breakthrough is crucial because improving the speed and efficiency of model optimisation allows scientists to explore a wider range of scenarios, from predicting the future state of ocean ecosystems to understanding past climate conditions. These models are key to projecting how much carbon the ocean can absorb as global temperatures rise, and how marine life, from plankton to fish populations, will respond.
“Climate models are only as good as their ability to capture the complexities of the ocean, which plays a vital role in regulating Earth’s climate,” Sophy explains. “A more efficient optimisation process allows researchers to fine-tune their models so they are closer to reality.”
Samar Khatiwala adds that “we’re currently exploring whether it is possible to reduce the spin-up time further, perhaps even down to a couple of hundred years, which would be transformational”. This ongoing research exploits a new acceleration technique developed by Dr. Khatiwala published recently in Science Advances. That study also used MITgcm to test the algorithm.
The team’s work holds promise for a wide range of applications in climate science, from enhancing predictions of harmful algal blooms to improving estimates of how much carbon the oceans can sequester. It also offers practical benefits for computational research, allowing labs to run larger simulations with fewer computational resources.
The full team comprises Coralia, Samar, Sophy Oliver from the National Oceanography Centre (and formerly a PhD student of Coralia's), Ben Ward from the University of Southampton and Iris Kriest from the GEOMAR Helmholtz Centre for Ocean Research Kiel, all supported by MITgcm. The work is published in JAMES (Journal of Modeling Earth Systems).
Many thanks to Helen Hill at MITgcm for allowing us to adapt their article and to Microsoft’s generative AI tool Copilot which they used to provide partial assistance in drafting a preliminary version of the text.