Accurately representing land-atmosphere interactions remains one of the key challenges in Earth system science. Process-based models, including land surface components of Earth System Models and dynamic global vegetation models, encode our understanding of these interactions and underpin projections of current and future climate. They also provide essential insight into carbon and water dynamics within ecosystems and help us extract richer information from our ever-growing observations. Yet advances in machine learning and artificial intelligence show that significant performance gains remain within reach. Harnessing these tools for modelling and understanding land-atmosphere interactions is of great potential to keep process-based models fit for purpose and at the cutting edge, supporting more robust climate projections. Robust benchmarking frameworks and rigorous comparisons are essential to evaluate, track and quality-control their applications.

This session invites contributions exploring how machine learning, artificial intelligence, and benchmarking can help address these challenges and accelerate model development. We welcome a broad range of perspectives, including: new AI-driven approaches to closing knowledge gaps; advances in benchmarking and model evaluation; physics-informed machine learning frameworks for operational use, process discovery and understanding; the growing role of emulators in model development and application; and the development of novel observational products and datasets. By bringing together researchers across disciplines, the session aims to highlight emerging opportunities for building more robust, reliable, and insightful land-surface models.