Model-Driven Machine Learning for Climate and Earth Science
The Model-Driven Machine Learning group aims to understand and predict the complex dynamics of Earth’s atmosphere, ocean, land and ice. Physics-based simulators handle this complexity well, but struggle with data assimilation, parameter tuning and uncertainty quantification. Machine learning thrives on large datasets, but ignores physics and generalizes poorly to new scenarios. We will combine physics- and data-driven approaches to gain insights unavailable to either approach alone. This work will serve to enhance our physical understanding of the Earth system and to improve predictions of high-impact events in the near and distant future.
We develop hybrid methods that combine the advantages of deep learning and physical modeling in a Bayesian framework. Examples of this hybrid approach include:
- Neural networks that solve differential equations.
- Algorithms that learn to infer model parameters using simulation results as training data.
- Machine learning models that respect physical laws.
- Flexible function approximators to fill gaps in our physical knowledge.
- Normalizing Flows, VAEs and GANs that model uncertainty in temperature, rainfall, fire and flooding.
The group is part of the Helmholtz AI initiative