HWRF HD33/H21A Comparison
Overview
The capabilities of the 2011 operational Hurricane Weather Research and Forecasting model (HWRF) have been ported to the Community HWRF code supported by the Developmental Testbed Center (DTC). A configuration of the community code (dubbed HD33) containing the 2011 operational capability was used to to create retrospective forecasts for the entire 2010 season in the North Atlantic and Eastern North Pacific basins. The forecasts were compared against pre-implementation tests of the HWRF operational model run at NCEP/EMC (dubbed H21A).
As explained in detail in the Test Plan, a match between HD33 and H21A was not expected due to differences between the HD33 and H21A configurations. In particular, HD33 was run in a Linux cluster, while H21A was run in an IBM P6. Indeed, the results indicate that while the intensity forecasts are similar between HD33 and H21A, the skill in track forecasts are different, with HD33 displaying larger skill in the Pacific, while H21A has more skill in the Atlantic.
These results highlight the large sensitivity of the 2011 operational HWRF to the computational platform. Diagnostic studies subsequent to this test and described in the Final Report have shed light on the causes for large platform sensitivity and code shortcomings have been addressed to reduce the dependency of the forecasts on platform in future HWRF runs.