The use of Atmospheric Motion Vectors (AMVs) in NWP (Numerical Weather Prediction) models continues to be an important source of information in data sparse regions. These AMVs are derived from a time-sequence of images from geostationary satellites and more recently from polar orbiting satellites. While NWP centers lament the fact that AMVs are difficult to work with, positive impact on model forecasts has been demonstrated. For example, the use of winds derived over the poles has shown forecast improvements not only in the polar regions, but also extending into the mid-latitudes.

Despite the success, AMVs are difficult to use for a some basic reasons:

1. Assimilation systems require error information for the measurements and that the spatial errors are uncorrelated. Currently, there is little error information associated with the vectors and there are spatially correlated errors inherent in the technique.

2. What does the AMV represent in terms of air motion? It has long been recognized that cloud motion does not necessarily represent air motion at a specific level. Nevertheless, tracking clouds and other features has been done since the first geostationary satellites were placed in orbit over 30 years ago.

There are several measures of the errors or quality of the AMVs, but each modeling center still applies unique thinning or blacklisting, many of which are empirically based. The goal is to better understand the errors in AMVs and provide estimates of these errors for both the vector and the assigned height, thereby producing wind sets that require less screening at the NWP sites.