A New Paradigm for the Derivation of LEO AMVs Without Wind Guess
In the field of Atmospheric Motion Vectors (AMVs), Low-Earth Orbit (LEO) platforms allow to collect wind observations from polar areas, which is not possible with sensors on board geostationary (GEO) platforms. Considering the longer time gaps between the sensing of matching atmospheric features in the LEO case, stronger deformations and displacements can occur, making the cross-correlation tracking yield poor results. For this reason, the derivation of LEO AMVs has always relied on the use of the wind guess. The guess is an estimation of the wind coming from the forecast model, and acts as a first approximation of the AMV, which is then refined by maximising the cross-correlation in a neighbourhood of the guess vector’s end location. The inherent problem behind the use of the wind guess is the dependency of the AMVs on the models they are meant to improve, with the risk of replicating and amplifying model errors. Additionally, our recent work shows the high sensitivity of the AMVs to any small error on the wind guess, highlighting the weak independence of AMV observations. For these reasons, EUMETSAT has designed and is currently testing a new tracking methods to derive LEO AMVs independently of the forecast. This method makes use of recent advancements in computer vision to allow the tracking of atmospheric features on long time gaps without wind guess. Our preliminary results on the derivation of AMVs from the Advanced Very-High-Resolution Radiometer (AVHRR) and the Sea and Land Surface Temperature Radiometer (SLSTR) suggest that this new method allows the derivation of AMVs of better quality (with statistical significance) than that of AMVs derived with wind guess and cross-correlation tracking.