Abstract: A hydrodynamic model of complex river network in which the computational cross-sections and reaches are fully decoupled is constructed for river network with dry beds and transition between supercritical and subcritical flows. The MUSCL-Hancock finite-volume scheme is adopted to discrete the Saint-Venant equations, and the method of Junction-Point Water Stage Prediction and Correction (JPWSPC) is used to calculate the hydrodynamic relations at junctions. The OpenMP and OpenACC applications are used to realize CPU-based and GPU-based parallel computing, respectively. Model accuracy is validated by benchmark cases with exact solutions, and the performance of parallel computing is demonstrated by real flood simulation of the Pearl River Delta. Results show that the proposed model is stable and effective, and is applicable for hydrodynamic simulation for steep river, thus has bright application prospects.

Key words: river network, hydrodynamics model, finite-volume method, GPU parallel computing, CPU parallel computing, numerical simulation