PNTS 2024 Theme Day – Artificial Intelligence and Physical Modelling in Earth Observation: Synergy or Competition?

The French National Remote Sensing Program(PNTS) is organizing its 12th Thematic Day. Artificial intelligence (AI) has been making great strides in many fields over the past few years, particularly in Earth system observation and modeling. Recently, some tasks usually performed by physical models, such as weather forecasting, have found themselves competing with AI-based systems (e.g. GraphCast). One of the strengths of AI algorithms is their great ability to learn tasks from large masses of data. Once an AI model has been “learned”, its application requires relatively few computing resources, making it very competitive with physical models, which are traditionally very expensive. Many AI applications make a direct link with physical modeling. For example, AI systems for weather forecasting are generally learned from ERA5 reanalyses, based on a physical model. These links between AI and physical modeling are not one-sided. AI systems can be influenced by physics, whether for learning, as in GraphCast, by constraints on the architecture (Physics-Informed Neural Networks – PINN), or by hybrid systems mixing AI parts with physical models. Conversely, AI systems can be analyzed to improve our understanding of physical processes (eXplainable AI – XAI). The aim of this PNTS day is to explore the links between AI and physical modeling, both in terms of their competing aspects, and also in terms of the various possible complementarities through applications linked to Earth observation by space remote sensing in the fields of continental surfaces, ocean physics and biogeochemistry, atmosphere, solid Earth, cryosphere and human sciences. The registration deadline for the PNTS theme day has been extended to Monday April 22, 2024. More information on the program and sessions on the official website.