Prediction–Correction Method for Parallelizing Implicit 2D Hydrodynamic Models. II: Application

Dechao Hu,Deyu Zhong,Yonghui Zhu,Guangqian Wang
DOI: https://doi.org/10.1061/(asce)hy.1943-7900.0001013
IF: 2.4
2015-01-01
Journal of Hydraulic Engineering
Abstract:A two-dimensional (2D) hydrodynamic model is developed for the Jing-Dongting river-lake system, with the computation domain (3,900 km(2)) covered by a channel-refined grid of 300,000 cells. The model provides good descriptions of the river-lake characteristics (strongly coupled, annular branches, varying flow regimes). In a simulation of a 1-year unsteady flow process, the mean absolute error in simulated water levels is 0.11-0.15 m, and the mean absolute relative error in simulated cross-section discharges is 4.3-8.5%, compared with field data. The accuracy of the 2D model is obviously better than those of existing one-dimensional (1D) and 1D-2D nested models. The prediction-correction parallelization (PCP) method is then tested by simulating the river-lake system using 64 subdomains. The complexities of real rivers are revealed to have almost no negative effects on the quasi-coupled solutions and accuracy of the PCP method. In tests, the mean absolute error in water level is found to be approximately 0.3 cm, and the mean absolute relative error in cross-section discharges is 0.23%, comparing the results of sequential and PCP simulations. In solving linear systems, sequential runs are 40.5 to 76.8 times slower than parallel runs using 64 cores. It takes 6.6 h to complete a simulation of a 1-year unsteady flow process in the river-lake system. The 2D model is then applied to the study of regulations of real river networks. (C) 2015 American Society of Civil Engineers.
What problem does this paper attempt to address?