696 High-Resolution Rapid Refresh Model Analytics in a High-Performance Computing Environment

Tuesday, 9 January 2018
Exhibit Hall 3 (ACC) (Austin, Texas)
Brian K. Blaylock, Univ. of Utah, Salt Lake City, UT; and J. D. Horel

The High Resolution Rapid Refresh (HRRR) forecast modeling system produces hourly analyses and 18 hr forecasts for the contiguous United States. Many people rely on these short-term forecasts for situational awareness and nowcasting. Long-term statistics of HRRR model output are of potential benefit for a variety of purposes.

A public archive of HRRR output beginning April 2015 is stored on the University of Utah’s Center for High Performance Computing Pando archive system. This highly efficient, object-storage archive allows access to the thousands of files to calculate seasonal and hourly statistics (e.g. percentiles, extremes, and forecast biases) of HRRR output variables. To reduce computational time, multiprocessing and multithreading methods are used to download multiple files from the archive simultaneously. For embarrassingly parallel computations, the Open Science Grid is being tested as a resource that can make synthesizing model statistics possible when traditional computing facilities have memory and computational limitations.

Among the many applications for HRRR composite statistics (e.g. renewable energy and agriculture), the focus of this work is to assist incident meteorologists who are responsible for giving weather forecasts to wild fire managers. Forecasters assigned to an incident can look at composite statistics to become familiar with numerical model performance for an area they may be unfamiliar with.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner