Source: Journal of Advances in Earth Systems Modeling (JAMES)
Weather and climate models have improved dramatically in recent years, as progress in one area has tended to benefit the other. But there is still significant uncertainty in the model outputs that are not precisely quantified. Indeed, the processes that govern climate and weather are chaotic, complex, and interconnected in ways researchers have yet to describe in the complex equations that power numerical models.
Historically, researchers have used approximations called parametrizations to model the relationships underlying fine-scale atmospheric processes and their interactions with large-scale atmospheric processes. Stochastic parameterizations have become increasingly common for representing uncertainty in subgrid-scale processes, and they are capable of producing fairly accurate weather forecasts and climate projections. But it is still a mathematically difficult method. Today, researchers are turning to machine learning to provide more efficiency to mathematical models.
Right here Gagne et al. evaluate the use of a class of machine learning networks known as generative adversarial networks (GANs) with a toy model of the extratropical atmosphere – a model first presented by Edward Lorenz in 1996 and thus known under the name of L96 system which has been frequently used as a testbed for stochastic parametrization schemes. The researchers trained 20 GANs, with varying noise amplitudes, and identified one set that outperformed a hand-tuned parameterization in L96. The authors found that the success of GANs in providing accurate weather forecasts was predictive of their performance in climate simulations: GANs that provided the most accurate weather forecasts also performed best for climate simulations, but they did not didn’t perform as well in offline ratings.
The study provides one of the first practically machine learning-relevant assessments for uncertain parameterizations. The authors conclude that GANs are a promising approach for the parameterization of small-scale but uncertain processes in weather and climate models. (Journal of Advances in Earth Systems Modeling (JAMES), https://doi.org/10.1029/2019MS001896, 2020)
—Kate Wheeling, science writer
Wheeling, K. (2020), Machine learning improves weather and climate models, Eos, 101, https://doi.org/10.1029/2020EO142422. Published on April 07, 2020.
Text © 2020. AGU. CC BY-NC-ND 3.0
Unless otherwise stated, images are subject to copyright. Any reuse without the express permission of the copyright holder is prohibited.