-
- H51V-0615: Deep Learning Atmospheric Models Reliably Simulate Out-of-Sample Temperature Extremes, and Blocking Frequencies
-
Board 0615‚ Hall EFG (Poster Hall)NOLA CC
Author(s):Generic 'disconnected' Message
Zilu Meng, University of Washington (First Author, Presenting Author)
Gregory Hakim, University of Washington
Wenchang Yang, Princeton University
Gabriel Vecchi, Princeton University
Understanding and predicting extreme weather events—like heatwaves, cold spells, and atmospheric blocking—is critical for preparing for climate impacts. Traditional atmospheric models are computationally demanding, which limits the study of extreme events, which are rare, over long periods of time. In this study, we test two newer atmospheric models that use deep learning: one fully based on data, and one that combines physical science with machine learning. We compare results from these models with a conventional high-resolution atmospheric model by simulating the weather 1900 to 2020. Our results show that both deep learning models can successfully reproduce the frequency of extreme events, even for the early 20th century, which was not used to train them. We also find that running these models, which are computationally very cheap, many times helps better capture climate variability. This research shows that combining machine learning with physics-based approaches offers a promising path toward better understanding and forecasting extreme weather in a changing climate.
Scientific DisciplineSuggested ItinerariesNeighborhoodType
Enter Note
Go to previous page in this tab
Session
