Transferring climate change knowledge

Accurate climate projections are required for climate adaptation and mitigation. Earth system model simulations, used to project climate change, inherently make approximations in their representation of small-scale physical processes, such as the formation of clouds, that are at the root of the uncertainties in global mean temperature's response to increased greenhouse gas concentrations. Several approaches have been developed to use historical observations to constrain future projections and reduce uncertainties in climate projections and climate feedbacks. Yet those methods cannot capture the non-linear complexity inherent in the climate system. Using a Transfer Learning approach, we show that Machine Learning, in particular Deep Neural Networks, can be used to optimally leverage and merge the knowledge gained from Earth system model simulations and historical observations to more accurately project global surface temperature fields in the 21st century. We reach a reduction in the 5-95% uncertainty range of global surface air temperature in 2081-2098 of up to 56% and 52% - across the Shared Socioeconomic Pathways considered - with respect to state-of-the-art approaches and the Sixth Assessment Report from the Intergovernmental Panel on Climate Change, respectively. We give evidence that our novel method provides narrower multi-model uncertainty together with more accurate climate projections, urgently required for climate adaptation.