I’m Mingshi Yang, a PhD student in Atmospheric Sciences at the University of Illinois Urbana-Champaign.

• mingshi3@illinois.edu
• Google Scholar

My research focuses on Arctic cyclones, including their dynamics, intensification mechanisms, predictability, and representation in both numerical and machine learning weather models.

  _{Image credit: NASA Earth Observatory}

• Cyclone climatology and storm-centric composite analysis using ERA5
  (Yang et al. 2024)
• Cyclone intensification mechanisms
  (Wang et al. 2024,
  Wang et al. 2026)
• Diagnostics of poleward energy transport
  (Yang et al. 2025)
• Analysis of impactful Arctic cyclone cases
  (Yang et al. 2026)
• Mechanism-dependent predictability and numerical / ML model evaluation
  (Yang et al. 2026, under review)

• Characteristics of extratropical cyclones under climate change in a global storm-resolving model
  (Yang et al. 2026, Accepted by JAMES)

• Applications of NeuralGCM (contributed)

  • Hierarchical evaluation of model performance across spatial scales
    (Chen et al. 2026)
  • Analysis of tropical cyclone frequency changes under climate change
    (Henry et al. 2026, under review)

• Programming & Scientific Computing: Python (NumPy, SciPy, Xarray, Pandas, Dask, MetPy), shell scripting, Jupyter, NetCDF/Zarr data processing
• Development Tools: Git/GitHub, open-source project development and maintenance, profiling and performance optimization
• Research & Project Execution: research design, independent project execution, scientific analysis and interpretation, manuscript development, collaborative research, and presentation of research outcomes
• Weather & Climate Modeling: WRF, HYSPLIT, ERA5/ECMWF and CMIP datasets, cyclone tracking, storm-centered compositing, model output analysis
• Machine Learning: PyTorch, machine-learning weather model training and evaluation, reinforcement learning, process-oriented model diagnostics
• High-Performance Computing: Slurm/PBS workflows, parallel workflows and large-scale computing
• Visualization: Matplotlib, Cartopy, OriginLab

• Physics-based simulation implemented entirely in vanilla Minecraft datapacks
• Custom collision, spin, and friction modeling
• Designed for realism and playability

• Developed RL agents for complex environments with large action spaces from scratch
• Efficient training achieving strong performance on Botzone ladder with compact models
• Fight the Landlord / Dou Dizhu
• Amazons Chess

Open to collaboration in:

• Weather and climate modeling
• Machine learning for geoscience
• Other interesting topics