Equip Deep Learning with Physical Insights: Towards a Symbiotic Integration for Hydrologic Modeling

Abstract:

Considering the respective merits of physical approaches and artificial intelligence (AI) models, the synergy of the two paradigms has recently been envisioned as an attractive research topic in the hydrological community, yet actualizing the organic integration remains an open question in hydrology. This study aims to propose a general approach to improving AI hydrologic awareness wherein physical approaches such as conceptual hydrologic models are included as special recurrent neural layers in a deep learning architecture. The illustrative case of runoff modeling across the conterminous United States shows that after “learning” a hydrologic model, the AI system has enhanced prediction accuracy and good intelligence to deal with unfamiliar regions and infer unobserved processes. The potential of AI for in-depth information mining, in return, fills the knowledge gap existing in physical approaches. This study represents a firm step toward realizing the vision of tackling Earth system challenges by physics-AI integration and demonstrates the symbiotic integration is critical to the success of such approaches.