Abstract
The divergent barotropic model presented in Part I is used to investigate reduction of rms forecast errors by periodic updating with model-produced observations. Results show that an asymptotic error level is reached in about 2 days. This rapid adaptation reflects the initial balancing provided to the data at each update. Asymptotic rms forecast errors are increasing functions of both the updating period and the observation error, but the asymptotic error level is shown to be independent of the initial error. These results are in basic agreement with experiments carried out with various numerical models. Error reduction by statistically optimal assimilation of data is expected to yield results similar to those obtained in a previous study by the author.
