Tom Beucler is a postdoctoral scholar in atmospheric science affiliated with the University of California, Irvine and Columbia University. He is interested in atmospheric physics and environmental fluid dynamics, combining theory, numerical simulations and observational analyses to improve the understanding of meteorology and climate, and guide the development of operational models of storms and clouds.
Tom visited the meteorological institute in Munich from January 7th to February 22nd 2019. He collaborated with Stephan Rasp to improve the representation of convection and clouds in global climate models using deep learning algorithms. In Rasp et al. (2018), the authors shown that neural networks could capture the main features of atmospheric heating, moistening and cloud radiation produced by storm-resolving models at a fraction of the computational cost. However, many questions need to be addressed before neural networks can be used operationally in climate models, including:
- Can we predict cloud formation processes in their full complexity?
- Can mass and energy be conserved while preserving the realism of the neural network models?
- How can we make sure that the neural network model interacts with its host climate model in a stable way?
These questions motivated the international collaboration between LMU, Columbia University, and the University of California, Irvine.
To read more about Tom Beucler, visit his homepage.