Physics-Based Artificial Intelligence in Earth Observation скачать в хорошем качестве

Physics-Based Artificial Intelligence in Earth Observation

Physics-Based Artificial Intelligence in Earth Observation 10 December 2025 by Prof. Fabio Del Frate Tor Vergata University of Rome, Italy The webinar presented a comprehensive argument for integrating Physics-Informed Artificial Intelligence (AI) into Earth Observation, asserting that maintaining contact with fundamental physics is essential for designing robust and dependable models. It is suggested that AI should move beyond being a "black box" by employing techniques like loss functions that encode physical laws and by integrating physics layers within deep neural network architectures. A primary strategy involves using simulation models, such as Radiative Transfer Models, to solve the Forward Problem—predicting observations from known physical states—thereby generating synthetic, physics-consistent data for training purposes. This approach is critical for tackling the ill-posed Inverse Problem, which involves retrieving geophysical parameters from sensor measurements, where AI adds significant value by uncovering complex, non-linear relationships. The principles are demonstrated through detailed examples, including the estimation of seismic parameters from SAR data and the retrieval of atmospheric and vegetation parameters. Webinar sponsored by IEEE GRSS Modeling in Remote Sensing (MIRS) Technical Committee Learn more: https://www.grss-ieee.org/event/physi...