Author(s): Zhaoxi Li; Tiejian Li; Chen Chen; Weidong Li; Jiaye Li

Linked Author(s): Guangqian Wang, Chen Chen

Keywords: Digital watershed modelling; Artificial intelligence; Physical mechanism; Runoff

Abstract: Watershed is a geographical entity with a series of hydrological and geomorphological characteristics, and it is the basic unit for the study and management of rivers. The traditional hydrological model based on physical process is widely used in watershed hydrological process simulation. This model relies on a large number of known physical parameters of the watershed, which limits its application in areas with scarce hydrological data. Data-driven models such as machine learning are widely used in watershed hydrological process simulation. However, the structure and parameters of data-driven model lack the underlying physical foundation, which makes interpretability a difficult problem. In this study, a generic physical-artificial intelligence digital watershed modelling framework was developed based on a watershed hydrological model by introducing physical parameters from the model into neural network modelling. The modeling results of the Yellow River Basin show that the physical-artificial intelligence digital watershed model has higher prediction accuracy while ensuring the physical constraint of the model itself. This study reveals that the neural network embedded with physical mechanism has great research potential in the field of hydrological research.

DOI: https://doi.org/10.3850/iahr-hic2483430201-340

Year: 2024