Enabling Distributed, Event Based, High Resolution Storm Surge Modeling

High-resolution predictions of coastal storm surge, inundation, and wind waves generally require substantial computational resources, particularly in a real-time, event-based mode. Regional real-time guidance systems using the ADCIRC tide, storm surge, and wave model have been operating in several risk-prone areas of the eastern US and Gulf coasts for the past several years using the Adcirc Surge Guidance System (ASGS). ADCIRC uses triangular finite elements to represent regional topography, bathymetry, critical hydraulic structures, and land cover at high-resolution. Multiple ASGS instances may be running at any time on widely distributed computing resources, but publishing their output in a standardized way. The distributed nature of ASGS instances and ADCIRC's unstructured grid data require an effective way to discover, gather, organize and visualize the model results across multiple data servers.

We have developed a system to provide uniform access to ASGS instance real-time outputs. The system has four essential components: 1) THREDDS/OPeNDAP servers to host and serve ADCIRC output; 2) CF-UGRID, a NetCDF Climate and Forecast metadata convention extension for unstructured grids; 3) NCTOOLBOX, a MATLAB toolbox (developed partly in the SURA/IOOS Coastal Ocean Modeling Testbed) that accesses common data model datasets using NetCDF-Java as the data access layer; and 4) a "data grid" and simple catalog that presents the collection of model outputs as one unified collection.

We demonstrate this data grid approach with AdcircViz, a MATLAB-based GUI being developed as part of NOAA's Joint Hurricane Testbed 2013 program. AdcircViz is an end-user application for visualizing and analyzing storm surge and waves in the collection of ASGS instances. It retrieves the data collection catalog, uses the UGRID metadata descriptions to populate the GUI fields and data structures, and uses NCTOOLBOX to retrieve data fields only when needed. It provides a consistent display of model results independent of the model's implementation details, inter-comparison of time series output at user-selected locations, and analyses of ensembles of simulations. Any model output that conforms to UGRID can be handled by the same methods.