Better use of correlation information in AMV extraction scheme

The Height Assignment (HA) is currently the most difficult task in the Atmospheric Motion Vectors (AMV) extraction scheme. Several sources of error can be introduced at the height assignment step, but one of the main difficulties is to clearly identify the pixels that lead the tracking process in the tracer box, in order to select them for the HA calculation. A good pixel selection process should ensure a direct link is kept between the feature really tracked and the calculation of the height. The most common method sorts the coldest pixels in the target box and uses them to calculate the AMV height. However, recent work showed that some of the coldest pixels can have very small and/or negative contributions to the cross correlation process. Indeed, it is then proposed to use individual pixel contribution-to-cross-correlation-coefficient information in the pixel selection process, in order to get a closer link between the feature tracked and the HA. This paper will present and discuss the latest results that have been investigated at EUMETSAT, in order to improve this pixel selection process in AMV extraction algorithms