Master Thesis Opportunity
Application of Scientific Machine Learning for Reactive Groundwater Transport and Coupled Subsurface Processes
Hydrogeology Group, Institute of Geosciences, University of Bonn
In geoscience, many important processes take place underground, including the transport of contaminants in groundwater, the movement of nutrients and carbon, and chemical reactions between water, minerals, and gases. Understanding and predicting these processes is essential for addressing challenges related to water quality, climate change, energy storage, and environmental protection.
Traditionally, scientists use complex numerical models to simulate these subsurface processes. While these models can be highly accurate, they often require large amounts of computing time, especially when chemical reactions and heterogeneous subsurface conditions are involved. Recent developments in Artificial Intelligence (AI) offer exciting new possibilities. In particular, a new class of machine learning methods called Neural Operators can learn the behavior of physical systems directly from simulation data and make predictions much faster than traditional numerical models.
In the hydrogeology work group, we offer a MSc thesis project, aiming to investigate how Scientific Machine Learning (SciML), particularly Neural Operators such as DeepONet, can be used to predict groundwater flow, solute transport, and reactive processes in subsurface environments. A particular focus will be on developing SciML-based models that can simultaneously predict multiple interacting variables, such as dissolved chemical species, gases, water-rock interactions, and other environmental quantities. The thesis will be a part of two ongoing interconnected research projects (ML-Mining and InterpAI-CoM), funded by Research Council of Finland and Jane and Aatos Erkko Foundation, on SciML and next generation AI methods for geoscience applications. The thesis will also involve close collaboration with the Geological Survey of Finland (GTK)
The project will combine:
- Generation and analysis of numerical simulation data from forward reactive transport simulations,
- Development of machine learning models using Python and PyTorch,
- Application of SciML techniques to environmental and geoscientific problems with an emphasis on groundwater transport,
- Evaluation of model accuracy, efficiency, and generalization capabilities
The specific topic may include but not limited to:
- SciML-based surrogate models for groundwater and reactive transport simulations,
- Physics-informed and physics-aware machine learning,
- DeepONet and Neural Operator architectures (Fig. 1),
- Multi-physics and multi-component environmental systems. Special attention will be given to challenging settings involving discontinuous initial conditions, heterogeneous reactive regions, multiscale transport structures, and partially observed physical processes
- Earth system and environmental process modeling.
Depending on your interests, the specific applications can also be tailored for any other systems including geophysics, geodynamics, mineralogy/geochemistry. The key point is to focus on coupled system of PDEs.
Basic requirements:
- Basic knowledge of groundwater flow, transport processes, or environmental modeling,
- Basic programming experience (preferably Python, but other programming experience is also welcome),
- Interest in Artificial Intelligence, Machine Learning, and data-driven modeling,
- Motivation to learn new computational and numerical methods,
- Ability to work independently and engage with scientific literature.
What you will gain:
- Hands-on experience with modern SciML methods for scientific applications
- Training in Python, PyTorch, and machine learning workflows
- Experience with numerical simulations and environmental modeling
- Exposure to cutting-edge research at the intersection of Geosciences and Artificial Intelligence
- Opportunities to contribute to ongoing research projects and potentially scientific publications
