The Met Office 3DVAR data assimilation scheme requires knowledge of the background error correlations in order to produce an analysis. Estimation of these errors is currently achieved by 24 hour and 48 hour forecast differences averaged over a two week period and so fails to give an indication of the daily variation in the magnitude and structure of the errors.

A new project at the Met Office uses an Error Breeding System and is being trialed as a route to better estimate the background 'Errors of the Day' (EotD). It provides errors as three-dimensional 'bredmodes' which originate as forecast differences between low-resolution short-range forecast run from perturbed analyses. Given a suitable bredmode structure, the EotD scheme allows the distribution of the observation within VAR to change according to the local bredmode structure, so that observational information can have a greater or lesser influence depending on the inferred local background error.

In this talk, the effect of bredmodes on the observational distribution and on the development of the subsequent forecasts will be discussed.