Investigating the impact of rapid cycling in a global data assimilation system

Most operational weather prediction centers cycle their global data assimilation systems at a cadence of two to four times per day for global weather prediction. Increases in computing power, observation availability, and horizontal model resolution motivate the development of a more rapidly-updated global assimilation system. Experiments using a hybrid 4DEnVar with 1-hour assimilation windows are performed in a version of the NOAA Global Data Assimilation System, and results are compared to those from traditional 6-hourly cycling. Skill is evaluated via comparisons with in situ observations, satellite winds, and regional analyses. Ensemble mean forecasts initialized from the hourly-cycled fields show significantly better skill for some variables and regions than those initialized from traditional 6-hourly-cycled fields, though they are informed by the same observations. These benefits appear only when dense, accurate observations are assimilated (e.g. aircraft wind). When aircraft data are withheld from assimilation, hourly cycling degrades the fit to in situ observations. Investigations into kinetic energy spectra demonstrate that fields from the hourly-cycling system retain more energy at synoptic and sub-synoptic scales than fields from the 6-hourly-cycling system. These results suggest that a global hourly-cycling assimilation system can use temporally and spatially dense observations more effectively than a 6-hourly-cycling system.