This paper documents development of a multiple‐Graphics Processing Unit (GPU) version of FUNWAVE‐Total Variation Diminishing (TVD), an open‐source model for solving the fully nonlinear Boussinesq wave equations using a high‐order TVD solver. The numerical schemes of FUNWAVE‐TVD, including Cartesian and spherical coordinates, are rewritten using CUDA Fortran, with inter‐GPU communication facilitated by the Message Passing Interface. Since FUNWAVE‐TVD involves the discretization of high‐order dispersive derivatives, the on‐chip shared memory is utilized to reduce global memory access. To further optimize performance, the batched tridiagonal solver is scheduled simultaneously in multiple‐GPU streams, which can reduce the GPU execution time by 20–30%. The GPU version is validated through a benchmark test for wave runup on a complex shoreline geometry, as well as a basin‐scale tsunami simulation of the 2011 Tohoku‐oki event. Efficiency evaluation shows that, in comparison with the CPU version running at a 36‐core HPC node, speedup ratios of 4–7 and above 10 can be observed for single‐ and double‐GPU runs, respectively. The performance metrics of multiple‐GPU implementation needs to be further evaluated when appropriate.
11 February 2021 in Journal of Geophysical Research: Solid Earth
A Probabilistic Tsunami Hazard Assessment (PTHA) study on the Chinese Mainland and the Taiwan Island was conducted. Characterized by broad and shallow continental shelves, the offshore region along Chinese Mainland's east coast yields a significant nonlinear effect and bottom friction to the propagating long waves. To address these shallow‐water effects, a fully nonlinear Boussinesq model was used in the computation‐based PTHA framework. We found that the inappropriate usage of the linear wave model could considerably overestimate the tsunami hazards along the East China Sea (ECS) and the Yellow Sea. Tsunami hazard along the coastline of Chinese Mainland is generally moderate. Elevated tsunami hazard levels are found along both flanks of the Yangtze and Pearl estuaries. The probability of these coastlines impacted by 1 m or greater tsunami waves (at 10‐m isobath) is about 14%–40% in next century. The shallow and tongue‐shaped submarine terrain amplifies the hazard level by trapping the tsunami energy in these areas. Major subduction zones in the Northwest Pacific were identified as the main sources of destructive events along the coast of ECS and the Taiwan Island, while the Manila Trench is the main source zone that threatens the Northern South China Sea. The tsunami hazards generated by the crustal earthquakes are modest, yet not negligible, particularly in the Taiwan Strait. We found the risk of tsunami inundation along the coast of Shanghai is low based on the hazard curves of total water level that incorporates the aleatory uncertainty of tides.
15 November 2023 in Natural Hazards and Earth System Sciences
Drownings due to rip currents are a major threat to beach safety. In this study a high-resolution Boussinesq model with a modified wave-resolving Lagrangian tracking module has been applied to a 2 km long embayed beach, Dadonghai of Sanya, Hainan Island, with the purpose of studying rip current variability, real-time rip hazard identification, and the optimal swimmer escape strategies. The beach stage evolves periodically at the study site and plays an important role in the long-term modulation of the occurrence and strength of rip currents according to the modeling. A series of tests are designed and confirm that rip current strength is closely related to wave properties and tidal levels. Spectral analysis of output time series at specific points shows that the modeled rip currents fluctuate on the orders of 1 and 10 min, which suggests the effects of wave-group-forced infragravity (IG) and very-low-frequency (VLF) motions. Rip hazard levels are defined by combining rip strength and its duration. An attempt to use the GPU-accelerated FUNWAVE-TVD (Total Variation Diminishing version of the Fully Nonlinear Boussinesq Wave Model) embedded with the spectral wave model WAM6-GPU (GPU version of the third-generation spectral wave model WAM Cycle 6) exhibits its capability to evaluate rip hazard levels in real time. One of the differences of the present study from previous works is that the random, wave-resolving tracking of virtual swimmers is performed with 1 m resolution to study beach safety strategies. The results demonstrate that multiple factors contribute to the survival of swimmers caught in the rip currents, including surf-zone bathymetry, rip strength, fine-scale flow patterns, the bather's position, and swimming ability. For weak-to-moderate rip currents and longshore currents, swim onshore consistently seems to be the most successful strategy across all the scenarios in this study. Higher surf-zone exit rates along Dadonghai beach are not favorable for stay afloat action, which puts swimmers at a higher risk of being expelled to deeper water. The effects of wave randomness of incoming wave trains and assignment of wave-following coefficients on Lagrangian tracking are also discussed.
The spectral wave model WAM (Cycle 6) is a commonly used code package for ocean wave forecasting. However, it is still a challenge to include it into the long-term Earth system modeling due to the huge computing requirement. In this study, we have successfully developed a GPU-accelerated version of the WAM model that can run all its computing-demanding components on GPUs, with a significant performance increase compared with its original CPU version. The power of GPU computing has been unleashed through substantial efforts of code refactoring, which reduces the computing time of a 7 d global 1/10° wave modeling to only 7.6 min in a single-node server installed with eight NVIDIA A100 GPUs. Speedup comparisons exhibit that running the WAM6 with eight cards can achieve the maximum speedup ratio of 37 over the dual-socket CPU node with two Intel Xeon 6236 CPUs. The study provides an approach to energy-efficient computing for ocean wave modeling. A preliminary evaluation suggests that approximately 90 % of power can be saved.
I/O is often a performance bottleneck in global ocean circulation models with fine spatial resolution. In this paper, we present an asynchronous parallel I/O framework and demonstrate its efficacy in the Mass Conservation Ocean Model (MaCOM) as a case study. By largely reducing I/O operations in computing processes and overlapping output in I/O processes with computation in computing processes, this framework significantly improves the performance of the MaCOM. Through both reordering output data for maintaining data continuity and combining file access for reducing file operations, the I/O optimizing algorithms are provided to improve output bandwidth. In the case study of the MaCOM, the cost of output in I/O processes can be overlapped by up to 99% with computation in computing processes as decreasing output frequency. The 1D data output bandwidth with these I/O optimizing algorithms is 3.1 times faster than before optimization at 16 I/O worker processes. Compared to the synchronous parallel I/O framework, the overall performance of MaCOM is improved by 38.8% at 1024 computing processes for a 7-day global ocean forecast with 1 output every 2 h through the asynchronous parallel I/O framework presented in this paper.
2 January 2026 in Journal of Geophysical Research: Oceans