Data-Driven and Data-Assisted Modeling for Applications in Fluid Dynamics and Geophysics

Explore data-driven and data-assisted modeling techniques for fluid dynamics and geophysics applications in this conference talk from the Machine Learning for Climate KITP conference. Dive into the challenges of predicting chaotic dynamical systems and high-resolution forecasting. Examine the benefits of machine learning in fluid modeling and discover hybrid architectures that combine traditional numerical methods with data-driven approaches. Learn about computational costs, initial prototypes, and real-world applications in numerical weather prediction. Gain insights into multi-time step optimization and data assimilation techniques. Engage with technical questions and discussions on the potential of these innovative modeling approaches for advancing climate science and Earth system understanding.

Introduction
Modeling a fluid dynamical system
Outline
Datadriven ML models
Dynamics models
Predicting chaotic dynamical systems
High resolution forecasting
Results
Why Machine Learning
Dataassisted forecasting
Hybrid model
Computational cost
Machine learning
Fluid modeling vs image processing
Hybrid architecture
High resolution trajectory
Initial prototype
High resolution spectrum
RMS error curves
Visual results
Hybrid numerical weather prediction
Preprint
Conclusion
Questions
Technical questions
Real data assimilation
MLPD Hybrid
Amount of data
Multitime step optimization