Real-time predictions of the 2023–2024 climate conditions in the tropical Pacific using a purely data-driven Transformer model
image: The Niño3.4 SST anomalies during 2020–2024 for reanalyses from GODAS (black dotted line) and predictions made by statistical (green dotted lines) and dynamic (blue dotted lines) models from the IRI collection and by the 3D-Geoformer (red dotted lines). Each colored line indicates the trajectory of a 9-month prediction made starting from different months; here, the data used in the statistical and dynamic models for the IRI are results taken directly from the IRI website at https://iri.columbia.edu/our-expertise/climate/forecasts/enso/current/. Note that this figure was produced in December 2023 when making real-time predictions that time; the related IRI results are available at http://iri.columbia.edu/our-expertise/climate/forecasts/enso/2023-December-quick-look
Credit: ©Science China Press
This study is led by Prof. Rong-Hua Zhang（School of Marine Sciences, Nanjing University of Information Science and Technology）. Following triple La Niña events during 2020–2022, the future evolution of climate conditions over the tropical Pacific has been a focused interest in ENSO-related communities. “ Current physics-based dynamical models exhibit large uncertainties and intermodal differences in real-time ENSO predictions,” Zhang says.
Zhang and his coworkers adopted a novel deep learning-based Transformer model to make real-time predictions for the 2023–2024 climate conditions in the tropical Pacific. The team found that the model can adequately depict the three-dimensional evolution of upper-ocean thermal anomalies and their interactions with sea surface temperature and surface winds.
The researchers also conduct sensitivity experiments to examine how prediction skills are affected by the input predictor specifications, including TIs during which information on initial conditions is retained for making predictions. A comparison with other dynamic coupled models is also made to demonstrate the prediction performance for the 2023–2024 El Niño event.
“These new exciting results indicate that the AI-based model, solely trained with data from the tropical Pacific and CMIP6 products, perform remarkably well in predicting ENSO, apparently surpassing other conventional statistical and dynamic models. Furthermore, the 3D-Geoformer can accomplish tasks that physical-driven models cannot achieve” Zhang says.
See the article:
Zhang R-H, Zhou L, Gao C, Tao L. 2024. Real-time predictions of the 2023–2024 climate conditions in the tropical Pacific using a purely data-driven Transformer model. Science China Earth Sciences, 67(12): 3709-3726, https://doi.org/10.1007/s11430-024-1396-x