Atmospheric data assimilation methods that estimate flow-dependent forecast statistics from ensembles are sensitive to sampling errors. This sensitivity is investigated in the context of vortex-scale hurricane data assimilation by cycling an ensemble Kalman filter to assimilate observations with a convection-permitting mesoscale model. In a set of numerical experiments, airborne Doppler radar observations are assimilated for Hurricane Katrina (2005) using an ensemble size that ranges from 30 to 300 members, and a varying degree of covariance inflation through relaxation to the prior. The range of ensemble sizes is shown to produce variations in posterior storm structure that persist for days in deterministic forecasts, with the most substantial differences appearing in the vortex outer-core wind and pressure fields. Ensembles with 60 or more members converge toward similar axisymmetric and asymmetric inner-core solutions by the end of the cycling, while producing qualitatively similar sample correlations between the state variables. Though covariance relaxation has little impact on model variables far from the observations, the structure of the inner-core vortex can benefit from a more optimal tuning of the relaxation coefficient. Results from this study provide insight into how sampling errors may affect the performance of an ensemble hurricane data assimilation system during cycling.
All Science Journal Classification (ASJC) codes
- Atmospheric Science