135 An Assessment of the S2S Forecast Skill of a Hybrid Model That Combines Machine Learning with an AGCM.

Monday, 29 January 2024
Hall E (The Baltimore Convention Center)
Dhruvit Patel, University of Maryland, College Park, MD

Handout (961.4 kB)

Several recent machine learning (ML) global weather forecasting models have demonstrated forecast skill comparable to state-of-the-art physics-based numerical weather prediction models at the short- and medium-range timescales [1-3]. However, the task of extending the range of useful forecasts using data-driven methods to the Subseasonal-to-Seasonal (S2S) timescales remains a challenge. Recent works [4-5] have trained ML methods to predict specific phenomena on the S2S timescales (e.g., El Nino Southern Oscillation), and on these tasks such models have outperformed state-of-the-art physics-based dynamical models. The use of such models is, however, limited to the specific task they were trained to perform and, in most cases, over a restricted geographic domain. We report on the climate forecasting skill of our hybrid global circulation model on the S2S timescales. Our model consists of a low-resolution atmospheric circulation model (SPEEDY) combined with ML components for the atmosphere and the ocean [6]. We demonstrate that our model can produce climate forecasts on the S2S timescales with useful skill and that our model’s forecasts capture various phenomena on these timescales (e.g., ENSO-precipitation correlations, MJO). The realistic correlations in our model’s outputs between oceanic variables and atmospheric variables indicate that our hybrid model has learned the coupled dynamics between the ocean and the atmosphere.

[1] Pathak J., Subramanian S., Harrington P., Raja S., Chattopadhyay A., Mardani M., Kurth T., Hall D., Li Z., Azizzadenesheli K., Hassanzadeh P., Kashinath K., and Anandkumar A., “FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators”, arXiv:2202.11214 (2022).

[2] Bi K., Xie L., Zhang H., Chen X., Gu X., and Tian Q, “Accurate medium-range global weather forecasting with 3D neural networks”, Nature 619, 533-538 (2023).

[3] Lam R., Sanchez-Gonzalez A., Willson M., Wirnsberger P., Fortunato M., Alet F., Ravuri S., Ewalds T., Eaton-Rosen Z., Hu W., Merose A., Hoyer S., Holland G., Vinyals O., Stott J., Pritzel A., Mohamed S., and Battaglia P., “GraphCast: Learning skillful medium-range global weather forecasting”, arXiv:2212.12794 (2022).

[4] Zhou L., and Zhang R.-H., “A self-attention-based neural network for three-dimensional multivariate modeling and its skillful ENSO predictions,” Science Advances 9 (2023).

[5] Ham Y.-G., Kim J.-H., and Luo J.-J., “Deep learning for multi-year ENSO forecasts”, Nature 573, 568-572 (2019).

[6] Arcomano T., Szunyogh I., Wikner A., Hunt B., and Ott E., “A Hybrid Atmospheric Model Incorporating Machine Learning Can Capture Dynamical Processes Not Captured by Its Physics-Based Component”, Geophysical Research Letters, 50 (2023).

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner