Abstract Process‐based hydrologic simulations in large domains generally require intensive computing resources. In this study, we implement various parallelization approaches within a process‐based hydrologic solver, SUMMA, including the Message Passing Interface (MPI), Open Multi‐Processing (OMP), and the Actor Model, to enable high‐performance computing for large‐domain hydrologic simulations. We provide detailed guidelines on these implementations to assist hydrologists in parallelizing their models effectively. Using a hydrologic simulation over North America as a case study, we compare the scalability, computational cost, input/output performance, and coupling capabilities of these parallel approaches with the original sequential approach. Our results show that the SUMMA‐MPI exhibits linear scaling up to 1,024 cores, whereas SUMMA‐OMP is only recommended for smaller numbers of cores. The MPI approach exhibited a straggler effect, resulting in core utilization of only 80%. To address this, we introduced a load‐balancing calibration based on historical runs, which increases SUMMA‐MPI core usage to 95% and thereby mitigates the straggler effect. With regard to coupling capabilities, MPI is the most effective for large‐scale simulations involving multiple nodes and extensive core counts, supporting strong coupling and synchronization. The Actor Model reveals its excellent fault tolerance that enables automatic modification and recommencement of specific Grouped Response Units (GRUs) rather than restarting the entire simulation in the event of a failure within the simulation. Through this study, the implementation details of multiple parallelization schemes are documented and their advantages and limitations are discussed, which provides parallel computing insights for advancing computational hydrology in the Earth System Science community.
Guo et al. (Thu,) studied this question.