A Deep Learning Parameterization of Gravity Wave Drag Coupled to an Atmospheric Global Climate Model - Aditi Sheshadri

Explore a deep learning parameterization of gravity wave drag coupled to an atmospheric global climate model in this conference talk from the Machine Learning for Climate KITP conference. Delve into the challenges of predicting future climate changes at regional and local scales due to complex multi-scale processes. Discover how big data and machine learning algorithms offer new opportunities to gain detailed insights into climate systems. Learn about the WaveNet model architecture and its application in predicting the Quasi-Biennial Oscillation (QBO). Examine the model's performance, including its stability over 100-year simulations and ability to generalize under increased CO2 scenarios. Investigate future directions, including Project Loon's potential for unprecedented coverage in gravity wave observation and analysis.

Intro
Gravity waves are ubiquitous..
Big source of uncertainty in climate prediction
Data Wave
Proof of concept.. Idealized model test
Model architecture - WaveNet
WaveNet does pretty well!
Predicting the QBO
Learning' one year is sufficient
Stable for 100 y times when run online
Increased CO2-WaveNet generalizes
QBO period, amplitude decrease
Future directions
Project Loon
Loon provides unprecedented coverage
Localize and analyze packets of gravity waves in time
Horizontal winds dominant predictors