AI models for water flow need real data to work, not just physics equations

A new study shows that AI models designed to simulate water flow, like rivers or floods, often fail if they only use physics equations. This means engineers and hydrologists cannot rely on physics alone; they must integrate real-world measurements to make these AI tools reliable.

Engineers and urban planners use complex simulations to predict floods, manage rivers, and design coastal defenses. The idea of 'physics-informed' AI was that it could build these models without much real-world data. This paper shows that for critical water systems, that idea was wrong: physics-only models often produce nonsense. It means that reliable AI models for these systems will always need some real-world measurements, even if sparse.