Lawrence Jing Yueh Liu

Welcome to my website! I am currently a PhD Student at the Department of Climate, Meteorology, and Atmospheric Science (CLIMAS) in University of Illinois Urbana-Champaign, working with Prof. Kelvin Droegemeier. I am interested in understanding and improving the predictability of deep convective storms, by combining physics-based numerical weather prediction (NWP) and machine learning approaches.

I started my research career at National Central University, where I studied data assimilation with Prof. Shu-Chih Yang. My master’s research focused on how large-scale environmental flow adjustment affects mesoscale precipitation prediction, using a multi-scale ensemble data assimilation system (WRF+LETKF). The key finding was that more accurate environmental flow leads to significantly better precipitation forecasts. This naturally raises further questions: How do different scales interact in a convective system? Is there a fundamental limit to how well we can predict a storm, and where does it come from?

Now at University of Illinois, I am exploring these questions in the context of deep convective storms. Traditional NWP models like WRF are physically rigorous but computationally expensive, while emerging AI/ML models like GraphCast and WoFSCast are fast but their physical fidelity at the storm scale remains largely unverified. My current research sits at this intersection: I use idealized simulations as a controlled testbed to evaluate whether AI/ML emulators truly capture convective dynamics rather than just reproducing statistical patterns. In parallel, I am also exploring machine learning approaches to data assimilation, with the goal of making storm-scale analysis more efficient. Can an AI model reproduce the physics of a supercell, not just its statistics? What does it even mean for a machine learning forecast to be physically consistent? I find these questions genuinely fascinating and worth pursuing.