Quantum Machine Learning for Climate Science

Abstract

Earth system models are fundamental to understanding and projecting climate change, although there are considerable biases and uncertainties in their projections. 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. The long-standing deficiencies in cloud parameterizations have motivated developments of global high-resolution cloud-resolving models that can explicitly resolve clouds and convection. Short simulations from the computationally costly high-resolution models together with observations can serve as information to develop machine learning (ML)-based parameterizations that are then incorporated into Earth system models, which is the goal of various current projects such as the USMILE ERC project. The KLIM-QML project, which is a project of the DLR/BMWK Quantum Computing Initiative, explores how quantum computing, and specifically quantum machine learning (QML) could be used to build upon recent progress in improving climate models using machine learning. Quantum machine learning models have shown remarkable expressive power and generalization capabilities, and are promising alternatives to classical machine learning in certain tasks. Our aim is thus to use QML to improve the representation of subgrid-scale phenomena in the ICOsahedral Non-hydrostatic (ICON) model. ICON is an open-access modelling framework, which is used on a variety of timescales and resolutions, ranging from numerical weather predictions to climate projections. We use regional and global cloud-resolving ICON simulations with data-driven techniques to train QML-based parametrizations, and evaluate their performance against commonly used parametrization schemes as well as machine learning models.