This paper introduces a machine learning approach to accelerate climate change projections by predicting long-term temperature responses from short-term simulations. Using a dataset of climate model simulations (HadGEM3), the study develops surrogate models (Ridge regression and Gaussian Process Regression) to map short-term to long-term temperature patterns. The results demonstrate the potential of this data-driven approach to capture regional patterns and diversity in climate responses, particularly for aerosol forcing scenarios, surpassing traditional pattern scaling methods. The study also highlights the importance of data sharing to improve the accuracy and applicability of these machine learning models.

L. A. Mansfield, P. J. Nowack, M. Kasoar, R. G. Everitt, W. J. Collins, A. Voulgarakis