Jing-Yi Zhuo

Jing-Yi Zhuo

Postdoc Research Scientist

Research

My research sits at the intersection of climate dynamics, tropical meteorology, and machine learning. I am broadly interested in understanding how the climate system responds to natural and anthropogenic forcing, with a focus on extreme weather events — particularly tropical cyclones.

Climate Model Biases and SST Patterns

A persistent challenge in climate science is that coupled climate models often misrepresent sea surface temperature (SST) patterns — most notably the “double-ITCZ” bias and the associated warm bias in the eastern Pacific cold tongue. My work investigates how these biases distort the simulated climate response to external forcing (e.g., aerosol reductions, Antarctic ozone depletion), and develops flux adjustment techniques to correct them.

Key questions: How do SST pattern biases shape the response to greenhouse gas forcing and aerosol changes? Can we use flux adjustment to produce more realistic climate simulations?

Related projects:

  • Flux adjustment of CESM2 → cesm2-fa
  • Eastern Pacific cooling under aerosol reduction (Under review)
  • Muted eastern Pacific cooling linked to double-ITCZ bias (In prep)

Tropical Cyclone Activity under Climate Change

Understanding how tropical cyclone (TC) frequency, intensity, and hazard respond to both natural variability and long-term warming is central to climate risk assessment. I study how SST warming patterns — whether from model bias or real-world forcing — modulate TC activity, and re-examine historical TC frequency trends using improved observational methods.

Key questions: How do SST warming patterns affect TC hazard? What do the historical trends in TC frequency really look like when observational heterogeneities are accounted for?

Related projects:

Physics-Informed Machine Learning for Tropical Cyclones

Satellite imagery offers a continuous, global record of TC structure, but translating raw imagery into reliable intensity and size estimates remains difficult. I develop deep learning models that incorporate physical constraints to improve these estimates, and apply them to build long, homogeneous datasets of TC inner-to-outer size.

Physics-augmented deep learning architecture for TC intensity and size estimation
Fig. 1 from Zhuo & Tan (2021, MWR): The physics-augmented deep learning framework — satellite IR imagery is fed into shared CNN layers, with physical auxiliary information injected at the task-specific layers for both single-task and multi-task learning.

Key questions: How can we encode physical knowledge into neural networks to improve TC monitoring? What do multi-decadal records of TC size reveal about long-term trends?

Related projects: