A superobbing scheme for atmospheric motion vectors

One of the primary limitations in assimilating geostationary atmospheric motion vectors (AMVs) in numerical weather prediction (NWP) models is the spatially correlated error inherent in the vectors. Correlated error violates a fundamental assumption in variational analysis, the assimilation technique used by most NWP centers. In order to reduce the effect of correlated error, most centers thin the data, thus discarding the majority of it. An alternative technique is to average the difference between the observation and the background, or innovation. This method of averaging the innovations, known as superobbing, reduces the number of assimilated winds as thinning does. There is, however, an additional benefit. Through averaging, there is also the potential to reduce some of the random error in the data. In this way superobbing is able to take advantage of the high resolution of many AMV datasets. This talk will describe some of the theory behind our superobbing scheme. It will also show the experimental results of superobbing trials on the Met Office global Unified Model (UM).