AI and the World's Hard Problems · Earth & climate science · arXiv · 2026-10-20
On the Adversarial Robustness of Hydrological Models
Hydrological models need to be trustworthy for operational water management, so the authors analyze how small, targeted perturbations to meteorological inputs affect simulated discharge. Using the fast gradient sign method across 1,347 German catchments, they compare a physical-conceptual model (HBV) with LSTM deep learning models. Catastrophic failure was rare, and LSTMs were generally more robust than HBV.