A dynamical atmospheric model combined with machine learning (ML) components capable of both accurate weather and climate simulations is an exciting hybrid dynamic-ML approach that has potential to substantially push the boundaries of current weather forecasting and climate prediction capabilities, according to a scientist at the ECI.
The new atmospheric model, named NeuralGCM, is presented in the journal Nature this week. The model outperforms some existing weather and climate prediction models and has the potential to make large savings in computational cost over conventional dynamic models.
General circulation models (GCMs), representing physical processes of the atmosphere, ocean, sea ice and land, are the basis for weather and climate predictions. Reducing the uncertainty around long-term prediction and estimating extreme weather events are key to helping understand climate mitigation and adaptation. Machine learning models and components have been suggested as an alternative approach to and enhancements of weather prediction with the benefit of reduced computational costs, but they often do not perform as well as GCMs when it comes to long-term prediction.
NeuralGCM combines physics-based core with machine learning methods, which can make medium-range weather forecasts as well as simulating climate over a number of decades. This hybrid model can compete with the accuracy of 1–15-day forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF develops one of the best conventional physics-based weather models). Also, NeuralGCM typically exceeds the accuracy of purely machine learning models.
Dr Neven Fuckar, Researcher at the Environmental Change Institute at the University of Oxford, said:
NeuralGCM is a state-of-the-art hybrid atmospheric model which shows higher skill than primarily data-driven atmospheric models based on ML architecture, and comparable or higher skill than the state-of-the-art dynamical models at higher resolution for medium-range (up to 10 days) weather forecasts to decadal climate predictions.
However, this novel hybrid model is much faster than typical dynamical counterparts of similar skill which allows about 1,000 to 100,000 times savings in computational time. This is exciting hybrid dynamic ML approach that has potential to substantially push the boundaries of current weather forecasting and climate prediction capabilities using already available or near-future planned computational and data storage resources in research and operational institutions around the world.”
NeuralGCM produces climate simulations at comparable level of accuracy as the best physics-based models. When the authors prescribed sea surface temperatures and sea ice concentration for 40-year climate projections using NeuralGCM, they found that the model well reproduced the global warming trends. NeuralGCM also surprisingly well predicts the trajectories of tropical cyclones. Together, these findings suggest that machine learning methods represent a viable approach for improving GCMs and reducing computational cost, the authors conclude.
A team of scientists from the USA and UK designed the model and authored the paper: Neural general circulation models for weather and climate