Assimilation Variability in the GEOS-1 Data Assimilation System

Abstract

Ensembles of assimilation runs were used to assess the sensitivity of the GEOS-1 (Goddard Earth Observing System?Version 1) data assimilation system to data gaps and changes in initial conditions. Perturbations from a ?control? assimilation were induced by withholding data for periods ranging from 12 to 96 h. Data assimilation then proceeded with each ensemble member for periods up to one month, and ensemble members (?assimilations?) were examined for convergence to the control assimilation. Experimental results show that this method is effective in identifying assimilation system weaknesses by determining where assimilations do not converge quickly. The methodology is also useful for determining assimilation ?spinup? time. For the GEOS-1 system, convergence of the assimilation ensemble was slow near the poles and in the Southern Hemisphere. This slow convergence was largely due to the sparseness of data in the Southern Hemisphere and to strong polar filtering. The differences between assimilations were primarily differences in the location or intensity of small-scale waves in the larger-scale flow, which tracked eastward with the movement of the small-scale waves. A ?fixed? quality control experiment showed that differences in quality control decisions contributed to maintaining differences between the ensemble members. The assimilation convergence was improved when a later version of the GEOS system, with weaker polar filtering, was used.