A novel paradigm for integrating physics-based numerical and machine learning models: A case study of eco-hydrological model. (May 2023)

• Record Type: Journal Article
• Authors: Chen, Chong; Zhang, Hui; Shi, Wenxuan; Zhang, Wei; Xue, Yaru
• Is Part Of: Environmental modelling & software. Volume 163(2023)
• Publication Date: 2023-05
• Subjects: Model integration -- Physics-based model -- Machine learning -- Eco-hydrological; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2023.105669
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Modeling is an essential tool for studying environmental systems. Due to the interdisciplinary nature, several models should be integrated to consider the nexus between different natural processes. However, the computational cost and accuracy of physics-based numerical models are always limited, even contradictory. In this study, a novel paradigm for integrating numerical models and machine learning models is proposed based on a previous study which integrates WOFOST, HYDRUS and MODFLOW. In the paradigm, the MODFLOW is replaced by Radial Basis Function neural network to reduce the computational costs with comparable accuracy. Object Modeling System is used as modeling framework which provides a run-time environment for the components. A synthetic case is implemented with data collected from Linze Inland River Basin Research Station in northwestern of China. Furthermore, the computation efficiency and generalization ability are discussed by refining the MODFLOW grids and changing the driving forces (precipitation). The proposed paradigm shows significant better performance on flexibility and computational efficiency with similar accuracy and acceptable generalization ability which would be further improved by larger dataset. This paper explores a computational efficiency paradigm by integrating physics-based and machine learning models which also provides a reference for physics-informed/guided model integration. Highlights: A new paradigm is proposed to integrate machine learning models and physics-based numerical models. WOFOST, HYDRUS and MODFLOW (later replaced by RBF neural network) are integrated in OMS. Two discretization schemes for the aquifer are implemented to demonstrate the computational efficiency. Several experiments on the driving forces are implemented to demonstrate the generalization ability. The new paradigm performs better in computational efficiency with comparable accuracy and generalization ability.