Harnessing Machine Learning and Physical Models for Sustainability

Sustainable development is a crucial concern in the modern world, as environmental challenges threaten the planet's ecological balance. In this context, machine learning has emerged as a potent tool for analyzing vast amounts of data to understand complex systems, making it crucial for addressing sustainability issues. However, data-driven models heavily rely on the quantity and quality of available labeled data. They also struggle to generalize beyond their training datasets. Additionally, these AI models often lack explainability, limiting their scientific usage as they may not always adhere to known laws of physics, such as the conservation of mass or energy. In contrast, physical models are grounded in scientific principles and can easily explain the relationship between input and output variables. However, they may struggle to extract information directly from data, and their simplicity might overlook important parameters. Furthermore, physical models typically have coarse resolution and are confined to specific time intervals and regions. By leveraging the wealth of environmental data alongside machine learning algorithms and knowledge-based models, this talk will explore how hybrid models can support sustainable environmental monitoring, including efficient disaster management, polar ice monitoring, and sea-level-rise uncertainty reduction.

Speaker: Maryam Rahnemoonfar, Lehigh University

Attend in person or online

Room: 350/372