Machine Learning for Ocean Closures - Advances and Lessons - Laure Zanna - Climate-C21

Explore machine learning applications for ocean closures in climate modeling through this conference talk. Dive into the challenges of informing society about future climate changes at regional and local scales. Examine how big data and machine learning algorithms can provide detailed insights into climate systems. Discover the potential for descriptive inference to drive new theories and validate existing ones. Learn about collaborative efforts to address key problems in climate science using data-driven approaches. Investigate the integration of machine learning with modeling experiments and parameterizations. Gain insights into current understanding and open questions in climate science, setting the stage for future research. Follow the speaker's journey through climate dynamics, mathematical models, and ocean modeling techniques. Explore methods for learning subgrid and turbulent closures, as well as the challenges of retraining models and handling strong extremes. Examine the implications of boundary conditions, missing forcings, and field of view on predictions. Delve into online modeling, stochastic parameters, and global kinetic energy considerations. Investigate probabilistic learning approaches and the potential for learning new physics equations. Conclude with a summary of key findings and engage in a Q&A session covering vertical fluxes and decomposers.

Intro
Climate Dynamicism
Grid Size
Mathematical Models
Current Climate Models
Why are we here
What are we doing
Ocean Model
Finding a Filter
Methods
Learning Subgrid Closures
Learning Turbulent Closures
The Good News
Retraining the Model
Strong Extremes
Boundary Conditions
Predictions vs Truth
Missing forcing
Field of view
Online Model
Stochastic Parameters
Global Kinetic Energy
Learning probabilistically
Learning equations
Learning new physics
Kinetic energy
Summary
Questions
Zoom
Vertical Fluxes
Decomposers