Abstract A previously developed research-grade (e.g. high-resolution) unstructured mesh of the northern Gulf of Mexico (named NGOM3) is optimized to produce a computationally efficient forecast-grade mesh for deployment in a real-time hurricane storm surge early warning system. The real-time mesh is developed from a mesh decimation scheme with focus on the coastal floodplain. The mesh decimation scheme reduces mesh nodes and elements from the research-grade mesh while preserving the representation of the bare-earth topography. The resulting real-time unstructured mesh (named NGOM-RT) contains 64% less mesh nodes than the research-grade mesh. Comparison of (ADCIRC + SWAN) simulated times-series and peak water levels to observations between the research-grade and real-time-grade meshes for Hurricanes Ivan (2004), Dennis (2005), Katrina (2005), and Isaac (2012) show virtually no difference. Model simulations with the NGOM-RT mesh are 1.5–2.0 times faster than using NGOM3 on the same number of compute cores.

中文翻译：

---

用于实时飓风风暴潮预测的非结构化有限元网格抽取

摘要 墨西哥湾北部先前开发的研究级（例如高分辨率）非结构化网格（命名为 NGOM3）经过优化，可生成计算效率高的预报级网格，用于部署在实时飓风风暴潮预警系统中。实时网格是从网格抽取方案发展而来的，重点是沿海漫滩。网格抽取方案减少了研究级网格中的网格节点和元素，同时保留了裸地地形的表示。生成的实时非结构化网格（名为 NGOM-RT）包含的网格节点比研究级网格少 64%。(ADCIRC + SWAN) 模拟时间序列和峰值水位与飓风 Ivan (2004)、Dennis (2005)、Katrina (2005) 的研究级和实时级网格之间的观测值的比较，和 Isaac (2012) 显示几乎没有区别。使用 NGOM-RT 网格进行模型模拟比在相同数量的计算核心上使用 NGOM3 快 1.5-2.0 倍。