Improving Tropical Climate Projections: From Numerical Modeling to Machine Learning

Abstract

Reliable climate projections are essential for adaptation and resilience in the tropics, yet current models struggle to capture the meso-scale atmosphere processes driving these events. This talk presents an integrated approach that bridges high-resolution numerical modeling and machine learning to improve both the fidelity and efficiency of tropical climate projections. I will first show how systematic calibration and bias-corrected dynamical downscaling enhance the realism of tropical simulations, and introduce CERA (Climate-Invariant Encoding through Representation Alignment), a physics-guided machine learning framework that generalizes across warming scenarios. Together, these advances pave the way for physics-consistent machine learning climate models that enable more credible and actionable projections of extreme weather in a changing climate.

Speaker Biography

Dr Shuchang Liu is a postdoctoral researcher at the Massachusetts Institute of Technology, funded by a fellowship from the Swiss National Science Foundation. She earned her PhD in Atmospheric and Climate Science from ETH Zürich, following a Master's at Tsinghua University and a Bachelor's at Nanjing University. Her research focuses on improving tropical climate projections by integrating high-resolution climate modelling and machine-learning methods.