Specific Humidity at the bottom model level as simulated by the CAMulator, an auto-regressive ML emulator of the Community Atmosphere Model (CAM6) that predicts the next atmospheric state from key inputs like sea surface temperatures, solar radiation, moisture, and atmospheric energy. It remains stable over long simulations and runs 350× faster than CAM6, enabling rapid generation of large ensembles.
 

At NSF NCAR, we are harnessing new ML technologies to expand our capabilities in initialized forecasting. One such effort includes participation in the AI Weather Quest, an international AI/ML-based real-time subseasonal forecasting competition. Our submission uses CAMulator, modified for real-time forecasting, to produce subseasonal predictions from 341 possible realizations of Earth’s future. As an ML-based emulator of the NSF NCAR Community Atmosphere Model (CAM), we can generate a real-time subseasonal forecast in under 3 minutes and a 20-year hindcast suite in 12 hours. This hindcast database will be made available to the community as a resource for subseasonal research to help continue improving our national resilience.

subCESMulator-logo.png


To further improve our research capabilities for S2S prediction, a new ML-based emulator is being developed. Situated between initial-value and boundary-condition predictability regimes, S2S timescales incorporate information from multiple components of the Earth system. Consequently, exploring subseasonal predictability requires tools that represent the coupled Earth system. At NSF NCAR, we are training subCESMulator, a new ML emulator specifically designed for S2S prediction that incorporates atmospheric, land, and ocean variables. We use CESM2-Large Ensemble output as training data. We are also developing a CESM-derived AI-ready dataset with additional high-frequency output for emulator training, compatible with the NSF NCAR CREDIT framework. More details on this new generation of ML-based emulation will be posted as they become available.

For more information or for partnership opportunities, please contact Kirsten Mayer.