The proposed approach integrates hybrid models that combine Earth system models with machine learning techniques and enables faster and more accurate simulations, essential for the design of global and regional climate policies. These models combine physics with AI, and improve generalizability in climate projections, as well as more accurately represent crucial Earth system processes.
The work is the result of collaboration between the principal investigators of the EuropeanResearch Council (ERC) Synergy Grant project 'Understanding and Modelling the Earth System with Machine Learning (USMILE)'. They are Professor Veronika Eyring and Professors Pierre Gentine (Columbia University, USA); Gustau Camps-Valls ( University of Valencia, Spain), Markus Reichstein (Max Planck Institute for Biogeochemistry, Germany) and David M. Lawrence (National Center for Atmospheric Research, USA).
“AI not only assists us, but is fundamental to redefine what our models can achieve”
“Integrating machine learning techniques with traditional climate modeling allows us to significantly advance our understanding of complex climate interactions and improve models. AI not only assists us, but is fundamental to redefine what our models can achieve,” says Gustau Camps-Valls, professor of Electronic Engineering and coordinator of Image and Signal Processing, a research group housed at the Image Processing Laboratory (IPL), located at the University of Valencia Science Park.
“The integration of AI into climate models represents a transformational step toward more accurate and useful projections,” says Veronika Eyring. This breakthrough addresses historical challenges in climate projection and improves the representation of small-scale processes and feedback mechanisms essential for climate dynamics. In addition, the publication marks a milestone in climate projection, with important implications for climate policy and strategies to reduce greenhouse gas emissions.