Machine learning for improved understanding and projections of climate change

Climate models are fundamental to understanding and projecting climate change. The models have continued to improve over the years, but considerable biases and uncertainties in their projections remain. A large contribution to this uncertainty stems from differences in the representation of phenomena such as clouds and convection that occur at scales smaller than the resolved model grid. This impacts the models’ ability to accurately project global and regional climate change, climate variability, and extremes. High-resolution, cloud resolving models with horizontal resolution of a few kilometers alleviate many biases of coarse-resolution models for deep clouds and convection, but
they cannot be run at climate timescales for multiple decades or longer due to high computational costs. In this talk I will present work from my group and collaborators where we use short regional as well as global cloud resolving
simulations to develop machine learning based atmospheric parametrizations for a reduction of long-standing biases in climate models. While unconstrained neural networks often learn spurious relationships that can lead to
instabilities in climate simulations, causally informed deep learning can mitigate this problem by identifying direct physical drivers of subgrid-scale processes. Trust and generalizability of the ML models can be further improved by
introducing physical constraints and equation discovery. Our approach can drive a paradigm shift in current climate and Earth system modelling towards a new data-driven, yet still physics-aware, ML-based hybrid climate model for
improved understanding and projections of climate change.