Quantum Computing and Quantum Machine Learning for Earth System Models

Abstract

Long-term climate projections require running global Earth system models on timescales of hundreds of years and have, due to their complexity and high computational costs, relatively coarse resolution (40-100 km in the horizontal direction). Unresolved sub-grid scale processes, such as clouds, convection or turbulence, are described in a statistical manner by so called parametrizations. Parametrizations are still one of the major sources of uncertainty in climate projections, owing to systematic model discrepancies as well as possibly sub optimal choices of the underlying free parameters. In the last years, there has been increasing focus on using machine learning (ML) methods to address these challenges. While enabling several improvements, the use of ML also highlights several requirements such as the ability to generalize to unseen physical scenarios, along with the need of robust and efficient multi-parameter optimization routines. Quantum computing offers alternative paradigms, which are promising candidates for addressing these challenges. Quantum machine learning (QML) models could benefit from better expressivity and generalization capability compared to classical ML, hence constituting promising Ansätze for accurate parametrizations in Earth system models. Quantum optimization algorithms such as quantum annealing are expected to provide substantial speed-ups compared to classical routines. They are thus suited as optimization subroutines in quantum-enhanced protocols for model tuning, a currently very timeconsuming step in the development of Earth system models. Here, we discuss these ideas and highlight opportunities and challenges in the use of quantum computers for improving climate and Earth system models and their analysis, and accelerating their development.