Transferring climate change physical knowledge

Kurzfassung

Precise and reliable climate projections are required for climate adaptation and mitiga-tion, but Earth system models still exhibit great uncertainties. Several approaches havebeen developed to reduce the spread of climate projections and feedbacks, yet thosemethods cannot capture the nonlinear complexity inherent in the climate system. Usinga Transfer Learning approach, we show that Machine Learning can be used to optimallyleverage and merge the knowledge gained from global temperature maps simulated byEarth system models and observed in the historical period to reduce the spread of globalsurface air temperature fields projected in the 21st century. We reach an uncertaintyreduction of more than 50% with respect to state-of- the-art approaches while givingevidence that our method provides improved regional temperature patterns togetherwith narrower projections uncertainty, urgently required for climate adaptation.