2025-07-23 –, Main Room 3
Traditional methods struggle to simulate pollutant transport in large oceanic domains. We propose a PINN framework integrating the advection-diffusion equation to predict pollution hotspots. Synthetic datasets with real-world variability and adaptive training strategies address coastal accumulation challenges. This approach supports scalable simulations, enabling future extensions like 3D modeling and real-time forecasting for environmental decision-making.
Ocean pollutant transport is a critical issue affecting marine ecosystems worldwide. The complexity of ocean currents and the variability of pollutant properties make it challenging to predict pollutant transport. This project develops a Julia-based model to simulate the transport of pollutants in ocean currents, leveraging Julia’s high-performance ecosystem—including DifferentialEquations.jl for solving PDEs, Flux.jl for neural network implementation, and Oceananigans.jl for ocean dynamics—to efficiently integrate physics and machine learning. This talk presents a novel approach to modeling ocean pollutant transport using Physics-Informed Neural Networks (PINNs).
We will begin by introducing the problem of ocean pollutant transport and the challenges associated with traditional simulation methods. We will then provide an overview of PINNs and their application to solving partial differential equations (PDEs), emphasizing how Julia’s SciML tools and Zygote.jl (for automatic differentiation) enable seamless integration of physical laws into neural networks. The core of the talk will focus on our PINN framework for modeling ocean pollutant transport, which integrates the advection-diffusion equation to predict pollution hotspots.
We will discuss the development of datasets that capture real-world variability and the implementation of training strategies to tackle the complex challenge of coastal accumulation, utilizing DataFrames.jl for preprocessing and CUDA.jl for GPU-accelerated training. Our results demonstrate the effectiveness of the PINN framework in simulating ocean pollutant transport, and we will highlight the potential for future extensions such as 3D modeling and real-time forecasting.
In conclusion, this talk will demonstrate the real-world implications of our research for environmental governance. By enabling dynamic forecasting capabilities, our PINN framework offers actionable insights to develop targeted strategies addressing oil spills and prioritizing remediation efforts. Ultimately, this work showcases the transformative potential of PINNs in oceanographic research, overcoming traditional modeling limitations to bridge the gap between theoretical research and real-world marine conservation challenges.
I am a DevOps Engineer with 7+ years of experience in IT, specializing in cloud infrastructure. My expertise includes automation, infrastructure as code, and building scalable system architectures, with a strong focus on optimizing cloud-native applications. I am keenly interested in AI/ML and enjoy working on projects integrating DevOps, AI, and large-scale simulations to drive innovation and efficiency.