Physics-Informed Deep Learning for Forecasting Coastal Dead Zones

NSF Project Overview and Funding

• Announcement Date: August 20, 2026.
• Lead Institution: NC State University.
• Funding Agency: U.S. National Science Foundation (NSF).
• Total Project Investment: A combined total of $1,049,875$ over three years (more than $1\text{ Million}$).
  • NC State Award: $800,013$.
  • Louisiana State University (LSU) Award: $249,862$.
• Formal Project Title: "Collaborative Research: CAIG: Physics-Informed Deep Learning for Understanding Coastal Hypoxia Formation Mechanisms."
• Primary Mission: To improve understanding and prediction of coastal hypoxia—marine areas with dangerously low dissolved oxygen levels, commonly known as "dead zones."
• Practical Application: Developing forecasting tools that run faster than traditional numerical models while remaining physically consistent, aimed at supporting fisheries management and decision-making across Gulf Coast communities.

Research Team Leadership and Collaboration

• Lead Principal Investigator (PI):
  • Paul Liu: Lead PI for the project and co-director of the AI Hub for Science at NC State.
• NC State University Team Members:
  • Ruoying He: Faculty Fellow at the AI Hub for Science.
  • Dongkuan "DK" Xu: Faculty Fellow at the AI Hub for Science.
• Louisiana State University (LSU) Team:
  • Kehui "Kevin" Xu: Professor at LSU leading the collaborating LSU team.

Hypoxia Formation Mechanisms and Environmental Dynamics

• Geographic Scope:
  • The largest coastal dead zone in the United States forms every summer along the Louisiana–Texas continental shelf.
• Environmental Drivers:
  • Nutrient Loading: High nutrient levels transported by the Mississippi River and Atchafalaya River stimulate intense biological production in coastal waters.
  • Physical Oceanography: Ocean circulation, water-column stratification (pycnocline formation separating the surface layer from the bottom layer), and seafloor benthic processes dictate where oxygen is depleted.
• Image Credit: NASA (Liam Gumley, Space Science and Engineering Center, University of Wisconsin-Madison and the MODIS science team).
• Scientific Challenge: Resolving how physical, chemical, and biological factors interact to determine the overall size, spatial distribution, and persistence of hypoxic zones.

Multi-Architecture Physics-Informed AI Framework

• Core Technical Architecture:
  • Multi-Architecture Physics-Informed Neural Network (PINN): Integrates physical constraints into deep learning algorithms.
  • Dynamic Graph Neural Networks (GNNs): Captures dynamic spatial-temporal relations across coastal grids.
  • Image-Segmentation Models: Analyzes dynamic spatial features and patterns in oceanographic data.
  • High-Resolution Ocean-Circulation Simulations: Provides baseline hydrodynamic input datasets.
  • Ocean Digital Twins: Offers high-fidelity virtual ocean environments for multi-scenario ocean modeling.
• Embedded Physical Constraints:
  • Water movement and physical transport.
  • Oxygen transport equations.
  • Light attenuation in coastal waters.
  • Sediment oxygen consumption at the seabed.
• Mathematical Governing Equations:
  • Fluid Momentum Equation:     $\frac{\partial u}{\partial t} + \mathcal{N}_u(u, p, c) = 0$
  • Oxygen Concentration Transport Equation:     $\frac{\partial c}{\partial t} + \nabla \cdot (u c) = \nabla \cdot (K \nabla c) + S$
  • Sediment Transport Equation:     $\frac{\partial s}{\partial t} + \nabla \cdot (u s) = \nabla \cdot (K_s \nabla s) - w_s \frac{\partial s}{\partial z} + E$
• Benthic and Sediment Dynamics Innovation:
  • Incorporates seabed-to-water oxygen exchange measurements.
  • Dynamic modeling of three sediment phases:
  1. Settling: Transport of particulate matter down to the seafloor.
  2. Resuspension: Re-entrainment of benthic sediments back into the water column (especially prominent during storm events).
  3. Benthic Oxygen Consumption: Microbial oxygen depletion within seabed sediments.
  • Helps isolate exact mechanisms driving how hypoxic waters form, spread, and persist.

Educational Impacts and Open Science Initiatives

• Interdisciplinary Student Training:
  • Educational and research training for graduate and undergraduate students spanning oceanography, coastal science, and artificial intelligence.
• Educational Platform:
  • Educational materials shared publicly via an AI-powered learning platform.
• Open Resources:
  • Project data, software tools, and trained AI models will be made available as open resources for the scientific community.