Author(s): Yuxuan Gao; Hao Hu; Dongfang Liang; Edoardo Borgomeo
Linked Author(s): Dongfang Liang
Keywords: Data-scarce catchments; Explainable AI; Hydrological prediction; Physics-informed neural networks; Transfer learning
Abstract: There is a lack of reliable hydrological prediction in many parts of the world due to insufficient streamflow gauging records [1]. Recent studies have applied transfer learning (TL), a machine learning technique that leverages information from data-rich (source) catchments to improve prediction accuracy in data-scarce (target) catchments. However, existing TL approaches are purely data-driven, which can lead to limited physical consistency and low interpretability [2, 3]. To address these challenges, we develop a physics-informed transfer learning (PITL) network. Hydrological principles are embedded into the network by using deep learning (DL) to automatically optimize key parameters of Hydrologiska Byråns Vattenbalansavdelning (HBV) hydrological model [4]. Since the network is fully differentiable, the physically meaningful parameters and DL parameters can be jointly updated during TL pre-training and fine-tuning stages.
Year: 2026