Investigation of Low Level AMV Height Assignment
Atmospheric Motion Vectors (AMVs) provide single-level wind estimates derived by tracking cloud features in image sequences from geostationary and polar orbiting satellites. They are established inputs to global as well as regional Numerical Weather Prediction (NWP) systems. Nevertheless, determining the heights of the winds as well as the assumption that clouds are passive tracers remain key sources of uncertainty in the use of AMVs. These aspects are often very difficult to examine, primarily due to a lack of independent wind observations over large parts of the oceanic regions. Recent work at ECMWF has focused on investigating possible height assignment issues, particularly inspired by a study in the Indian Ocean which highlighted challenging regions for the assimilation of low level AMVs in tropical inversion regions. These regions are often associated with relatively sharp changes in the wind speed in the vertical, leading to a particular sensitivity to AMV height assignment errors.
Here we present the results from investigation into errors in the height assignment of which the initial focus has been on the low level AMVs. Potential connections between the assigned heights and model cloud parameters are explored with the aim of identifying a systematic correction or improved quality control procedures. Analysis of background departure statistics (comparison of observations with the model background) showed that AMVs placed above the model cloud could potentially have a more detrimental impact than those placed unrealistically close to the surface. However, simply screening these observations resulted in negative impacts in assimilation experiments. Reassigning the AMV pressure using model cloud height information showed it was possible to achieve improvements in Root Mean Square Vector Difference (RMSVD) and speed bias. Assimilation experiments while showing positive impact in the verification against own analysis, showed mostly neutral changes in the fit of independent conventional wind observations to the model background. However, positive signals were seen in Aeolus and scatterometer winds and in microwave imagers which are primarily sensitive to changes in the cloud. Overall, reassignment to the cloud base or average pressure of the layer performed better than the cloud top in assimilation experiments. Combined with the results from the background departures, the option to reassign the heights of low level AMVs diagnosed above the model cloud to the average pressure of the cloud layer is recommended for future operational use at ECMWF.
