Development and Implementation of the NOAA National Water Model Using High Performance Computing Resource on the NSF/NCAR and NOAA Supercomputing Systems (Invited Presentation)

The operational NOAA National Water Model (NWM) was designed to provide real-time, regularly cycling analyses and forecasts of hydrologic conditions for the U.S. These analyses and forecasts from the NWM provide high resolution (sub 1km and river reach scale) descriptions of important water resource and water-related risk variables for the nation. The NWM takes advantage of multiple spatial framework and multiple time-stepping capabilities available within the community WRF-Hydro system and the NOAA NWM configuration of WRF-Hydro is the largest deployment of the system to date, in terms of spatial extent, computational, memory and local disk requirements. This talk describes the development timeline and testing of the NWM configuration of WRF-Hydro as the system was ‘scaled-up’ for implementation over the NWM modeling domain. First a timeline of development is presented which outlines both the initial requirements and main software development and implementation milestones. Results from computational benchmark tests are presented which document how certain modeling decisions and code changes resulted in significant computational savings in the operational cycling of the NWM. Benchmark results from both the NSF/NCAR ‘yellowstone’ supercomputer as well as NOAA’s WCOSS system are provided and preliminary benchmark results from the new NSF/NCAR ‘cheyenne’ system will be presented if possible. Specific modifications to the model’s parallelization and input/output (I/O) structures are described. The talk concludes with a summary of lessons learned and upcoming development activities related to the next version of the NWM.