In partial fulfillment of the requirements for the degree of

Doctor of Philosophy in Ocean Science & Engineering

In the School of Earth and Atmospheric Sciences

Qi Zhang

Will defend her dissertation

EARTH SYSTEM SIMULATIONS OF OCEAN CARBON AND OXYGEN: A WAY FORWARD WITH A MACHINE LEARNING EMULATOR 

March 31st, 2026 at 11 AM

 Ford ES&T, Room 3243 (The Ocean Room)

Zoom Link: https://gatech.zoom.us/j/99336477116

 Thesis Advisors:

Takamitsu Ito, Ph.D.

School of Earth and Atmospheric Sciences

Georgia Institute of Technology

Annalisa Bracco, Ph.D.

School of Earth and Atmospheric Sciences

Georgia Institute of Technology

Committee Members:

Joseph Montoya, Ph.D.

School of Biological Sciences

Georgia Institute of Technology

Kevin Haas, Ph.D.

School of Civil and Environmental Engineering

Georgia Institute of Technology

Ali Sarhadi, Ph.D.

School of Earth and Atmospheric Sciences

Georgia Institute of Technology

ABSTRACT:

The Atlantic Ocean plays a central role in Earth’s climate system by regulating heat, carbon and oxygen through circulation and biogeochemical processes. Understanding how these processes are represented in the Earth System Models (ESMs) is critical for assessing future climate and ocean biogeochemical change. This thesis examines three aspects of Atlantic biogeochemistry. First, it investigates the mechanisms impacting carbon uptake in the subpolar North Atlantic, showing that variability in the Atlantic Meridional Overturning Circulation (AMOC) influences regional carbon sequestration through changes in freshwater input, alkalinity and surface ocean carbonate chemistry. Second, it addresses large uncertainty in ESM simulated dissolved oxygen (O2), particularly in the representation of Atlantic Oxygen Minimum Zones (OMZs), where ESMs often disagree in both magnitude and trends. To reduce these biases, this work develops and applies a machine learning emulator, O2EMU, that learns empirical relationships between O2, temperature (T) and salinity (S) from observations and applies them to ESM simulations. The emulator substantially reduces inter-model spread and climatological biases in Atlantic O2 fields while reproducing observed patterns of OMZ variability and long-term deoxygenation. Third, extending this framework to climate scenario simulations reveals that while observationally constrained relationships can improve projections of O2 variability, uncertainties remain due to divergent projections of T and S among ESMs. Together, these results demonstrate how combining mechanistic understanding of carbon cycling with data-driven emulation of O2 can improve the representation and interpretation of Atlantic biogeochemistry in climate models and provide a pathway toward more reliable projections of future ocean O2 variability in the Atlantic and globally.