AI for Climate Science seminar series: Why We Still Need Traditional Climate and Earth System Models – and How They Might Be Improved

We are pleased to announce the second talk in the IIASA-wide AI Seminar Series, featuring Professor Steven Sherwood from the University of New South Wales, Sydney.

Professor Sherwood is a renowned climate scientist at UNSW's Climate Change Research Centre, where he leads pioneering work at the intersection of atmospheric physics, climate dynamics, and model development. His research focuses on improving our understanding of climate processes through both traditional modelling approaches and innovative techniques, including machine learning and high-resolution simulations. With a distinguished academic career spanning NASA, Yale University, and UNSW, he brings deep expertise to the challenges and opportunities in next-generation climate modelling.

For online participation, a registration is necessary.

Please register here.

Please note that the seminar is going to be recorded.

Title:
Why We Still Need Traditional Climate and Earth System Models – and How They Might Be Improved

Abstract:
Climate modelling has so far been done with models that only coarsely resolve processes, leading to errors, and predictions that can lack regional specificity. With the advent of global, high-resolution numerical models has come the potential for fine-scale predictions that overcome the limitations of traditional physics parameterisations required in the standard highly truncated models.  I will argue that traditional models do indeed have significant problems in simulating climate changes other than global-mean temperature, that are not confined to their lack of regional specificity but affect planetary-scale changes. These remain not fully appreciated, and require major efforts toward model improvement. I will however also argue that high-resolution models cannot replace them for many of the most important applications of climate modelling to inform mitigation and adaptation decisions. Focusing on the problem of atmospheric convection, and in particular the prediction of convective scale phenomena, I will outline some novel strategies including the use of new testing approaches and machine learning, and a broader range of numerical simulation strategies, that might show how our standard climate models can be fixed.